Skip to content

Codebase intelligence for AI and humans: code health scores, auto-generated docs, git analytics, dead code detection, and architectural decisions via MCP.

Add reasoning mode configuration (#175)

Eridanusbb380a5b166e
diff --git a/.env.example b/.env.example
index 7c8b4c04f..4b00e975c 100644
--- a/.env.example
+++ b/.env.example
@@ -72,6 +72,12 @@ REPOWISE_DEFAULT_MODEL=claude-sonnet-4-6
# Max concurrent LLM calls during generation
REPOWISE_MAX_CONCURRENT=5
+# Reasoning mode for supported docs-generation providers: auto | off | minimal
+# auto = provider default; off = disable Qwen3/OpenRouter effort reasoning when supported;
+# minimal = request the lowest supported OpenAI/OpenRouter reasoning effort.
+# Unsupported explicit modes fail before an API call.
+REPOWISE_REASONING=auto
+
# Max pages to regenerate per maintenance run
REPOWISE_CASCADE_BUDGET=30
diff --git a/README.md b/README.md
index 253f63662..e22e4f8ab 100644
--- a/README.md
+++ b/README.md
@@ -577,9 +577,10 @@ repowise reindex # rebuild vector store (no LLM calls)
`repowise init` generates `.repowise/config.yaml`. Key options:
```yaml
-provider: anthropic # anthropic | openai | ollama | litellm
+provider: anthropic # anthropic | openai | openrouter | gemini | deepseek | ollama | litellm | mock
model: claude-sonnet-4-5
embedding_model: voyage-3
+reasoning: auto # auto | off | minimal
git:
co_change_commit_limit: 500
@@ -594,6 +595,10 @@ maintenance:
background_regen_schedule: "0 2 * * *"
```
+`reasoning` applies to documentation generation. `auto` preserves provider
+defaults; explicit `off` / `minimal` modes are translated only by providers and
+models that support them, otherwise repowise fails before making an API call.
+
Full configuration reference: [docs/CONFIG.md](docs/CONFIG.md)
---
diff --git a/docs/CLI_REFERENCE.md b/docs/CLI_REFERENCE.md
index 4af4b81f1..1c50dcac9 100644
--- a/docs/CLI_REFERENCE.md
+++ b/docs/CLI_REFERENCE.md
@@ -50,7 +50,7 @@ In workspace mode, adds: repo scanning, per-repo indexing, cross-repo analysis (
| Flag | Description |
|------|-------------|
-| `--provider` | LLM provider: `anthropic`, `openai`, `gemini`, `ollama`, `mock` |
+| `--provider` | LLM provider: `anthropic`, `openai`, `openrouter`, `gemini`, `deepseek`, `ollama`, `litellm`, `mock` |
| `--model` | Model name override (e.g., `claude-sonnet-4-6`) |
| `--embedder` | Embedder for semantic search: `gemini`, `openai`, `mock` |
| `--index-only` | Skip LLM generation. Only parse, build graph, and index git. Free. |
@@ -61,6 +61,7 @@ In workspace mode, adds: repo scanning, per-repo indexing, cross-repo analysis (
| `--exclude / -x` | Gitignore-style exclusion patterns. Repeatable. |
| `--include-submodules` | Include git submodule directories. |
| `--concurrency` | Max concurrent LLM calls (default: 5). |
+| `--reasoning` | Reasoning mode for supported providers: `auto`, `off`, or `minimal` (default: `auto`). |
| `--resume` | Resume from the last checkpoint if interrupted. |
| `--force` | Regenerate all pages even if they exist. |
| `--commit-limit` | Max commits to analyze per file (default: 500). |
@@ -76,6 +77,8 @@ repowise init --provider anthropic --yes # automated
repowise init --index-only # free, no LLM
repowise init --dry-run # preview cost
repowise init --test-run # quick test (10 files)
+repowise init --provider openai --model qwen3 --reasoning off
+repowise init --provider openrouter --model openai/gpt-5 --reasoning minimal
repowise init -x vendor/ -x "*.gen.go" # exclude patterns
repowise init --include-submodules # include submodules
repowise init . # workspace mode
@@ -95,6 +98,7 @@ Incrementally update wiki pages for files changed since the last sync.
| `--provider` | Override LLM provider for this run |
| `--model` | Override model |
| `--since` | Git ref to diff from (overrides `state.json`) |
+| `--reasoning` | Reasoning mode for supported providers: `auto`, `off`, or `minimal` |
| `--cascade-budget` | Max pages to regenerate (default: auto) |
| `--dry-run` | Show what would be updated without regenerating |
| `--workspace` | Update all stale repos in the workspace + cross-repo analysis |
@@ -109,6 +113,7 @@ Incrementally update wiki pages for files changed since the last sync.
repowise update # diff since last sync
repowise update --dry-run # preview
repowise update --since v1.0.0 # diff from a tag
+repowise update --reasoning off # one-off supported-provider thinking-off run
repowise update --workspace # all workspace repos (incl. first-time indexing)
repowise update --repo backend # specific workspace repo
repowise update --no-workspace # force single-repo mode in a workspace root
diff --git a/docs/USER_GUIDE.md b/docs/USER_GUIDE.md
index 644a091da..053511f27 100644
--- a/docs/USER_GUIDE.md
+++ b/docs/USER_GUIDE.md
@@ -162,7 +162,7 @@ repowise init [PATH]
| Flag | Description |
|------|-------------|
-| `--provider` | LLM provider: `anthropic`, `openai`, `gemini`, `ollama`, `mock`. Auto-detected from env vars if not set. |
+| `--provider` | LLM provider: `anthropic`, `openai`, `openrouter`, `gemini`, `deepseek`, `ollama`, `litellm`, `mock`. Auto-detected from env vars if not set. |
| `--model` | Model name override (e.g., `claude-sonnet-4-6`, `gpt-5.4-nano`) |
| `--embedder` | Embedder for semantic search: `gemini`, `openai`, `mock`. Auto-detected from env vars. |
| `--index-only` | Skip LLM generation entirely. Only parse, build graph, and index git. Free. |
@@ -172,6 +172,7 @@ repowise init [PATH]
| `--skip-infra` | Exclude infrastructure files (Dockerfiles, Makefiles, Terraform, shell scripts). |
| `--exclude / -x` | Gitignore-style exclusion patterns. Repeatable: `-x vendor/ -x "*.generated.*"` |
| `--concurrency` | Max concurrent LLM calls (default: 5). Higher = faster but more API pressure. |
+| `--reasoning` | Reasoning mode for supported providers: `auto`, `off`, or `minimal` (default: `auto`). |
| `--resume` | Resume from the last checkpoint if a previous run was interrupted. |
| `--force` | Regenerate all pages even if they already exist. |
| `--commit-limit` | Max commits to analyze per file (default: 500, max: 5000). Saved to config. |
@@ -197,6 +198,9 @@ repowise init --provider openai --dry-run
# Quick test with 10 files
repowise init --provider gemini --test-run
+# OpenRouter with minimal reasoning effort
+repowise init --provider openrouter --model openai/gpt-5 --reasoning minimal
+
# Exclude vendor and generated code
repowise init -x vendor/ -x "*.gen.go" -x "**/__generated__/**"
```
@@ -228,6 +232,7 @@ Much faster and cheaper than a full `init` — only regenerates pages for change
| `--provider` | Override LLM provider for this run |
| `--model` | Override model |
| `--since` | Git ref to diff from (overrides `state.json`). Example: `--since v1.0.0` |
+| `--reasoning` | Reasoning mode for supported providers: `auto`, `off`, or `minimal`. |
| `--cascade-budget` | Max pages to regenerate per run (default: 30). Prevents runaway regeneration. |
| `--dry-run` | Show what would be updated without regenerating. |
@@ -246,6 +251,9 @@ repowise update --since v2.0.0
# Limit regeneration scope
repowise update --cascade-budget 10
+
+# Disable reasoning for a supported provider/model for this run
+repowise update --reasoning off
```
---
diff --git a/docs/architecture/ARCHITECTURE.md b/docs/architecture/ARCHITECTURE.md
index 7ddb56872..8a13e8394 100644
--- a/docs/architecture/ARCHITECTURE.md
+++ b/docs/architecture/ARCHITECTURE.md
@@ -1617,7 +1617,7 @@ Key files:
Full configuration with defaults (`.repowise/config.yaml`):
```yaml
-provider: anthropic # anthropic | openai | ollama | litellm
+provider: anthropic # anthropic | openai | openrouter | gemini | deepseek | ollama | litellm | mock
model: claude-sonnet-4-5 # passed through to the provider
embedding_provider: anthropic
embedding_model: voyage-3
@@ -1694,4 +1694,4 @@ rate_limits:
ollama:
rpm: 999999
tpm: 999999
-```
\ No newline at end of file
+```
diff --git a/packages/cli/src/repowise/cli/commands/init_cmd.py b/packages/cli/src/repowise/cli/commands/init_cmd.py
index c323056c9..1ec542de0 100644
--- a/packages/cli/src/repowise/cli/commands/init_cmd.py
+++ b/packages/cli/src/repowise/cli/commands/init_cmd.py
@@ -21,6 +21,7 @@ from repowise.cli.helpers import (
load_config,
load_state,
resolve_provider,
+ resolve_reasoning,
resolve_repo_path,
run_async,
save_config,
@@ -277,7 +278,9 @@ async def _persist_result(
# Record a completed GenerationJob so the web UI can show
# "last synced" / "last re-indexed" timestamps.
- from datetime import datetime, UTC as _UTC
+ from datetime import UTC as _UTC
+ from datetime import datetime
+
from repowise.core.persistence.crud import upsert_generation_job
now = datetime.now(_UTC)
@@ -318,6 +321,7 @@ def _run_workspace_generation(
skip_tests: bool,
skip_infra: bool,
test_run: bool,
+ reasoning: str = "auto",
) -> list[Any]:
"""Run LLM generation for a single repo in the workspace init flow.
@@ -326,20 +330,28 @@ def _run_workspace_generation(
workspace run.
"""
from repowise.cli.cost_estimator import build_generation_plan, estimate_cost
+ from repowise.cli.helpers import get_db_url_for_repo
from repowise.cli.ui import BRAND, RichProgressCallback
from repowise.core.generation import GenerationConfig
from repowise.core.generation.cost_tracker import CostTracker
- from repowise.core.persistence.vector_store import InMemoryVectorStore
- from repowise.core.providers.embedding.base import MockEmbedder
- from repowise.core.pipeline import run_generation
- from repowise.cli.helpers import get_db_url_for_repo
from repowise.core.persistence import (
create_engine as _ce,
+ )
+ from repowise.core.persistence import (
create_session_factory as _csf,
+ )
+ from repowise.core.persistence import (
get_session as _gs,
+ )
+ from repowise.core.persistence import (
init_db as _idb,
+ )
+ from repowise.core.persistence import (
upsert_repository as _ur,
)
+ from repowise.core.persistence.vector_store import InMemoryVectorStore
+ from repowise.core.pipeline import run_generation
+ from repowise.core.providers.embedding.base import MockEmbedder
# Build embedder
embedder_impl: Any
@@ -371,7 +383,10 @@ def _run_workspace_generation(
vector_store = InMemoryVectorStore(embedder_impl)
# Cost estimate
- gen_config = GenerationConfig(max_concurrency=concurrency)
+ gen_config = GenerationConfig(
+ max_concurrency=concurrency,
+ reasoning=resolve_reasoning(reasoning),
+ )
plans = build_generation_plan(
result.parsed_files, result.graph_builder, gen_config, skip_tests, skip_infra
)
@@ -475,6 +490,7 @@ def _workspace_init(
skip_infra: bool = False,
concurrency: int = 5,
test_run: bool = False,
+ reasoning: str | None = None,
yes: bool = False,
dry_run: bool = False,
resume: bool = False,
@@ -574,6 +590,7 @@ def _workspace_init(
if adv.get("exclude"):
exclude_patterns = list(exclude_patterns) + list(adv["exclude"])
test_run = adv.get("test_run", test_run)
+ reasoning = adv.get("reasoning") or reasoning
embedder_name_resolved = _resolve_embedder(adv.get("embedder") or embedder_name)
elif not index_only:
# "full" mode
@@ -582,6 +599,8 @@ def _workspace_init(
)
# Resolve provider once (shared across all repos for generation)
+ primary_cfg = load_config(primary_repo.path)
+ resolved_reasoning = resolve_reasoning(reasoning, primary_cfg)
provider = None
if not index_only:
try:
@@ -591,6 +610,8 @@ def _workspace_init(
f"Model: [cyan]{provider.model_name}[/cyan]"
)
console.print(f" Embedder: [cyan]{embedder_name_resolved}[/cyan]\n")
+ if resolved_reasoning != "auto":
+ console.print(f" Reasoning: [cyan]{resolved_reasoning}[/cyan]\n")
except Exception as exc:
console.print(f" [yellow]Provider setup failed ({exc}); falling back to index-only.[/yellow]")
index_only = True
@@ -693,6 +714,7 @@ def _workspace_init(
skip_tests=skip_tests,
skip_infra=skip_infra,
test_run=test_run,
+ reasoning=resolved_reasoning,
)
result.generated_pages = generated_pages
total_pages += len(generated_pages)
@@ -753,6 +775,7 @@ def _workspace_init(
embedder_name_resolved,
exclude_patterns=exclude_patterns if exclude_patterns else None,
commit_limit=resolved_commit_limit,
+ reasoning=resolved_reasoning,
)
# Save workspace config with updated timestamps
@@ -867,7 +890,10 @@ def _workspace_init(
"--provider",
"provider_name",
default=None,
- help="LLM provider name (anthropic, openai, gemini, deepseek, ollama, litellm, mock).",
+ help=(
+ "LLM provider name (anthropic, openai, openrouter, gemini, "
+ "deepseek, ollama, litellm, mock)."
+ ),
)
@click.option("--model", default=None, help="Model identifier override.")
@click.option(
@@ -888,6 +914,12 @@ def _workspace_init(
"--force", is_flag=True, default=False, help="Regenerate all pages, ignoring existing."
)
@click.option("--concurrency", type=int, default=5, help="Max concurrent LLM calls.")
+ "--reasoning",
+ type=click.Choice(["auto", "off", "minimal"]),
+ default=None,
+ help="Reasoning mode for supported providers: auto, off, or minimal. Default: auto.",
+)
@click.option(
"--test-run",
is_flag=True,
@@ -951,6 +983,7 @@ def init_command(
resume: bool,
force: bool,
concurrency: int,
+ reasoning: str | None,
test_run: bool,
index_only: bool,
exclude: tuple[str, ...],
@@ -1012,6 +1045,7 @@ def init_command(
skip_tests=skip_tests,
skip_infra=skip_infra,
concurrency=concurrency,
+ reasoning=reasoning,
test_run=test_run,
yes=yes,
dry_run=dry_run,
@@ -1073,6 +1107,7 @@ def init_command(
skip_tests = adv["skip_tests"]
skip_infra = adv["skip_infra"]
concurrency = adv["concurrency"]
+ reasoning = adv.get("reasoning") or reasoning
exclude = adv["exclude"]
test_run = adv["test_run"]
embedder_name = adv.get("embedder") or embedder_name
@@ -1090,6 +1125,7 @@ def init_command(
# Merge exclude_patterns from config.yaml and --exclude/-x flags
config = load_config(repo_path)
language = config.get("language", "en")
+ resolved_reasoning = resolve_reasoning(reasoning, config)
exclude_patterns: list[str] = list(config.get("exclude_patterns") or []) + list(exclude)
# Resolve commit limit: CLI flag → config.yaml → default (500)
@@ -1153,13 +1189,22 @@ def init_command(
console.print(f" Embedder: [cyan]{embedder_name_resolved}[/cyan]")
if language != "en":
console.print(f" Language: [cyan]{language}[/cyan]")
+ if resolved_reasoning != "auto":
+ console.print(f" Reasoning: [cyan]{resolved_reasoning}[/cyan]")
# Validate provider connection
from repowise.core.providers.llm.base import ProviderError
with console.status(" Verifying provider connection…", spinner="dots"):
try:
- run_async(provider.generate("You are a test.", "Reply with OK.", max_tokens=50))
+ run_async(
+ provider.generate(
+ "You are a test.",
+ "Reply with OK.",
+ max_tokens=50,
+ reasoning=resolved_reasoning,
+ )
+ )
except ProviderError as exc:
raise click.ClickException(f"Provider validation failed: {exc}") from exc
console.print(" [green]✓[/green] Provider connection verified")
@@ -1269,7 +1314,11 @@ def init_command(
# Cost estimation
from repowise.core.generation import GenerationConfig
- gen_config = GenerationConfig(max_concurrency=concurrency, language=language)
+ gen_config = GenerationConfig(
+ max_concurrency=concurrency,
+ language=language,
+ reasoning=resolved_reasoning,
+ )
plans = build_generation_plan(
result.parsed_files, result.graph_builder, gen_config, skip_tests, skip_infra
)
@@ -1365,13 +1414,21 @@ def init_command(
# is persisted to the llm_costs table. We need the repo_id from the
# database row that was created/upserted during _persist_result
# (which has not run yet), so we look it up or fall back to in-memory.
- from repowise.core.generation.cost_tracker import CostTracker
from repowise.cli.helpers import get_db_url_for_repo
+ from repowise.core.generation.cost_tracker import CostTracker
from repowise.core.persistence import (
create_engine as _create_engine,
+ )
+ from repowise.core.persistence import (
create_session_factory as _create_sf,
+ )
+ from repowise.core.persistence import (
get_session as _get_session,
+ )
+ from repowise.core.persistence import (
init_db as _init_db,
+ )
+ from repowise.core.persistence import (
upsert_repository as _upsert_repo,
)
@@ -1522,6 +1579,7 @@ def init_command(
embedder_name_resolved,
exclude_patterns=exclude_patterns if exclude_patterns else None,
commit_limit=resolved_commit_limit if commit_limit is not None else None,
+ reasoning=resolved_reasoning,
)
# ---- Completion panel ----
diff --git a/packages/cli/src/repowise/cli/commands/update_cmd.py b/packages/cli/src/repowise/cli/commands/update_cmd.py
index 9adb920c6..ee854fd3b 100644
--- a/packages/cli/src/repowise/cli/commands/update_cmd.py
+++ b/packages/cli/src/repowise/cli/commands/update_cmd.py
@@ -19,12 +19,12 @@ from repowise.cli.helpers import (
release_update_lock,
resolve_command_target,
resolve_provider,
+ resolve_reasoning,
resolve_repo_path,
run_async,
save_state,
)
-
# ---------------------------------------------------------------------------
# Mode resolution
# ---------------------------------------------------------------------------
@@ -173,6 +173,12 @@ def _workspace_update(
@click.option("--model", default=None, help="Model identifier override.")
@click.option("--since", default=None, help="Base git ref to diff from (overrides state).")
@click.option("--concurrency", type=int, default=5, help="Max concurrent LLM calls.")
+ "--reasoning",
+ type=click.Choice(["auto", "off", "minimal"]),
+ default=None,
+ help="Reasoning mode for supported providers: auto, off, or minimal.",
+)
@click.option(
"--cascade-budget",
type=int,
@@ -228,6 +234,7 @@ def update_command(
provider_name: str | None,
model: str | None,
since: str | None,
+ reasoning: str | None,
cascade_budget: int | None,
dry_run: bool,
workspace: bool,
@@ -353,7 +360,11 @@ def update_command(
cfg = load_config(repo_path)
language = cfg.get("language", "en")
- config = GenerationConfig(max_concurrency=concurrency, language=language)
+ config = GenerationConfig(
+ max_concurrency=concurrency,
+ language=language,
+ reasoning=resolve_reasoning(reasoning, cfg),
+ )
# Read exclude patterns from config (set during init or via web UI)
exclude_patterns: list[str] = list(cfg.get("exclude_patterns") or [])
@@ -692,7 +703,9 @@ def update_command(
# Record a GenerationJob so the web UI "last synced" timestamp updates
try:
- from datetime import datetime, UTC as _UTC
+ from datetime import UTC as _UTC
+ from datetime import datetime
+
from repowise.core.persistence.crud import upsert_generation_job
async with get_session(sf) as session:
diff --git a/packages/cli/src/repowise/cli/commands/workspace_cmd.py b/packages/cli/src/repowise/cli/commands/workspace_cmd.py
index b95eed4e6..909dd8ac2 100644
--- a/packages/cli/src/repowise/cli/commands/workspace_cmd.py
+++ b/packages/cli/src/repowise/cli/commands/workspace_cmd.py
@@ -3,6 +3,7 @@
from __future__ import annotations
from pathlib import Path
+from typing import TYPE_CHECKING
import click
from rich.table import Table
@@ -10,10 +11,14 @@ from rich.table import Table
from repowise.cli.helpers import (
console,
find_workspace_root,
+ resolve_reasoning,
resolve_repo_path,
run_async,
)
+if TYPE_CHECKING:
+ from repowise.core.workspace.config import WorkspaceConfig
+
# ---------------------------------------------------------------------------
# Helpers
@@ -493,6 +498,7 @@ def _run_index_for_repo(
# Run LLM doc generation through the existing single-repo init pathway
# so we get cost gating, cascading, and full parity with `repowise init`.
generated_pages = 0
+ resolved_reasoning = resolve_reasoning(config=primary_cfg)
if generate_docs and provider is not None:
try:
generated_pages = _generate_docs_for_added_repo(
@@ -500,6 +506,7 @@ def _run_index_for_repo(
provider=provider,
embedder_name=embedder_name,
concurrency=concurrency,
+ reasoning=resolved_reasoning,
exclude_patterns=exclude_patterns,
)
console.print(f" [green]✓[/green] Generated {generated_pages} pages")
@@ -529,6 +536,7 @@ def _run_index_for_repo(
embedder_name,
exclude_patterns=exclude_patterns or None,
commit_limit=int(commit_limit) if commit_limit else None,
+ reasoning=resolved_reasoning,
)
# Update workspace config entry
@@ -563,6 +571,7 @@ def _generate_docs_for_added_repo(
provider: object,
embedder_name: str,
concurrency: int,
+ reasoning: str,
exclude_patterns: list[str],
) -> int:
"""Generate wiki pages for a newly-added workspace repo.
@@ -613,7 +622,10 @@ def _generate_docs_for_added_repo(
continue
graph_builder.build()
- config = GenerationConfig(max_concurrency=concurrency)
+ config = GenerationConfig(
+ max_concurrency=concurrency,
+ reasoning=reasoning,
+ )
assembler = ContextAssembler(config)
generator = PageGenerator(provider, assembler, config, language=config.language)
diff --git a/packages/cli/src/repowise/cli/helpers.py b/packages/cli/src/repowise/cli/helpers.py
index 7dd4461d3..0b11e91ac 100644
--- a/packages/cli/src/repowise/cli/helpers.py
+++ b/packages/cli/src/repowise/cli/helpers.py
@@ -13,7 +13,13 @@ from typing import Any, Literal, TypeVar
import click
from rich.console import Console
-CONFIG_FILENAME = "config.yaml"
+from repowise.core.reasoning import (
+ ReasoningMode,
+)
+from repowise.core.reasoning import (
+ resolve_reasoning as resolve_core_reasoning,
+)
+from repowise.core.repo_config import CONFIG_FILENAME, load_repo_config
T = TypeVar("T")
@@ -223,22 +229,18 @@ def get_head_commit(repo_path: Path) -> str | None:
def load_config(repo_path: Path) -> dict[str, Any]:
"""Load ``.repowise/config.yaml`` or return an empty dict if absent."""
- config_path = get_repowise_dir(repo_path) / CONFIG_FILENAME
- if not config_path.exists():
- return {}
- text = config_path.read_text(encoding="utf-8")
- try:
- import yaml # type: ignore[import-untyped]
+ return load_repo_config(repo_path)
- return yaml.safe_load(text) or {}
- except ImportError:
- # Simple line-by-line parser for the flat key: value format we write
- result: dict[str, Any] = {}
- for line in text.splitlines():
- if ":" in line:
- k, _, v = line.partition(":")
- result[k.strip()] = v.strip()
- return result
+
+def resolve_reasoning(
+ reasoning: str | None = None,
+ config: dict[str, Any] | None = None,
+) -> ReasoningMode:
+ """Resolve generation reasoning from CLI flag, env, config, then default."""
+ try:
+ return resolve_core_reasoning(reasoning, config)
+ except ValueError as exc:
+ raise click.ClickException(str(exc)) from exc
def save_config(
@@ -249,6 +251,7 @@ def save_config(
*,
exclude_patterns: list[str] | None = None,
commit_limit: int | None = None,
+ reasoning: str | None = None,
) -> None:
"""Write provider/model/embedder (and optionally exclude_patterns) to ``.repowise/config.yaml``.
@@ -266,6 +269,8 @@ def save_config(
existing["exclude_patterns"] = exclude_patterns
if commit_limit is not None:
existing["commit_limit"] = commit_limit
+ if reasoning is not None:
+ existing["reasoning"] = resolve_reasoning(reasoning)
try:
import yaml # type: ignore[import-untyped]
@@ -277,6 +282,8 @@ def save_config(
except ImportError:
# Fallback: write simple key-value format (lists not supported)
lines = [f"provider: {provider}", f"model: {model}", f"embedder: {embedder}"]
+ if reasoning is not None:
+ lines.append(f"reasoning: {resolve_reasoning(reasoning)}")
config_path.write_text("\n".join(lines) + "\n", encoding="utf-8")
diff --git a/packages/cli/src/repowise/cli/ui.py b/packages/cli/src/repowise/cli/ui.py
index 16c8a7f50..57b7064aa 100644
--- a/packages/cli/src/repowise/cli/ui.py
+++ b/packages/cli/src/repowise/cli/ui.py
@@ -746,6 +746,12 @@ def interactive_advanced_config(
type=int,
)
+ result["reasoning"] = click.prompt(
+ " Reasoning mode",
+ default="auto",
+ type=click.Choice(["auto", "off", "minimal"]),
+ )
+
# Embedder selection
detected_embedder = _resolve_embedder_from_env()
embedder_choices = ["gemini", "openai", "mock"]
@@ -772,6 +778,7 @@ def interactive_advanced_config(
summary.add_row("Commit limit", str(result["commit_limit"]))
summary.add_row("Follow renames", "yes" if result["follow_renames"] else "no")
summary.add_row("Concurrency", str(result["concurrency"]))
+ summary.add_row("Reasoning", result["reasoning"])
summary.add_row("Embedder", result["embedder"])
if patterns:
if len(patterns) <= 5:
diff --git a/packages/core/src/repowise/core/generation/models.py b/packages/core/src/repowise/core/generation/models.py
index 4b2690eea..0b50458b4 100644
--- a/packages/core/src/repowise/core/generation/models.py
+++ b/packages/core/src/repowise/core/generation/models.py
@@ -14,6 +14,8 @@ from dataclasses import dataclass, field
from datetime import UTC, datetime
from typing import Literal
+from repowise.core.reasoning import ReasoningMode, normalize_reasoning
+
# ---------------------------------------------------------------------------
# PageType and generation levels
# ---------------------------------------------------------------------------
@@ -64,6 +66,7 @@ class GenerationConfig:
max_concurrency: asyncio.Semaphore size for parallel calls.
embed_concurrency: asyncio.Semaphore size for vector-store writes.
Defaults to max_concurrency.
+ reasoning: Provider-level reasoning intent.
cache_enabled: In-memory SHA256 prompt deduplication.
staleness_threshold_days: Days before a page is considered stale.
expiry_threshold_days: Days before a page is considered expired.
@@ -76,6 +79,7 @@ class GenerationConfig:
token_budget: int = 48000
max_concurrency: int = 5
embed_concurrency: int | None = None
+ reasoning: ReasoningMode = "auto"
cache_enabled: bool = True
staleness_threshold_days: int = 7
expiry_threshold_days: int = 30
@@ -90,6 +94,7 @@ class GenerationConfig:
def __post_init__(self) -> None:
if self.embed_concurrency is None:
object.__setattr__(self, "embed_concurrency", self.max_concurrency)
+ object.__setattr__(self, "reasoning", normalize_reasoning(self.reasoning))
# ---------------------------------------------------------------------------
diff --git a/packages/core/src/repowise/core/generation/page_generator.py b/packages/core/src/repowise/core/generation/page_generator.py
index 750956c19..9d97c47d8 100644
--- a/packages/core/src/repowise/core/generation/page_generator.py
+++ b/packages/core/src/repowise/core/generation/page_generator.py
@@ -968,6 +968,7 @@ class PageGenerator:
max_tokens=self._config.max_tokens,
temperature=self._config.temperature,
request_id=request_id,
+ reasoning=self._config.reasoning,
)
if self._config.cache_enabled:
diff --git a/packages/core/src/repowise/core/pipeline/orchestrator.py b/packages/core/src/repowise/core/pipeline/orchestrator.py
index 30fdfa8ab..a477c1d4d 100644
--- a/packages/core/src/repowise/core/pipeline/orchestrator.py
+++ b/packages/core/src/repowise/core/pipeline/orchestrator.py
@@ -166,6 +166,7 @@ async def run_pipeline(
resume: bool = False,
progress: ProgressCallback | None = None,
cost_tracker: Any | None = None,
+ generation_config: Any | None = None,
) -> PipelineResult:
"""Run the repowise indexing/analysis/generation pipeline.
@@ -200,6 +201,10 @@ async def run_pipeline(
Limit generation to top 10 files by PageRank (for quick validation).
progress:
Optional callback for progress reporting. Pass None for silent operation.
+ generation_config:
+ Optional ``GenerationConfig`` used for page generation. When omitted,
+ generation uses repo-local config such as ``reasoning`` from
+ ``.repowise/config.yaml`` and ``REPOWISE_REASONING``.
Returns
-------
@@ -322,6 +327,17 @@ async def run_pipeline(
if progress:
progress.on_message("info", "Phase 3: Generation")
+ resolved_generation_config = generation_config
+ if resolved_generation_config is None:
+ from repowise.core.generation import GenerationConfig
+ from repowise.core.reasoning import resolve_reasoning
+ from repowise.core.repo_config import load_repo_config
+
+ resolved_generation_config = GenerationConfig(
+ max_concurrency=concurrency,
+ reasoning=resolve_reasoning(config=load_repo_config(repo_path))
+ )
+
generated_pages = await run_generation(
repo_path=repo_path,
parsed_files=parsed_files,
@@ -335,6 +351,7 @@ async def run_pipeline(
concurrency=concurrency,
progress=progress,
resume=resume,
+ generation_config=resolved_generation_config,
)
# ---- Execution flow tracing -----------------------------------------------
diff --git a/packages/core/src/repowise/core/providers/llm/anthropic.py b/packages/core/src/repowise/core/providers/llm/anthropic.py
index bc188dcc7..bdc5ee386 100644
--- a/packages/core/src/repowise/core/providers/llm/anthropic.py
+++ b/packages/core/src/repowise/core/providers/llm/anthropic.py
@@ -13,18 +13,19 @@ Recommended models (as of 2026):
from __future__ import annotations
import os
+from typing import TYPE_CHECKING, Any, AsyncIterator
import structlog
+from anthropic import APIStatusError as _AnthropicAPIStatusError
from anthropic import AsyncAnthropic
from anthropic import RateLimitError as _AnthropicRateLimitError
-from anthropic import APIStatusError as _AnthropicAPIStatusError
from tenacity import (
+ RetryError,
+ before_sleep_log,
retry,
retry_if_exception_type,
stop_after_attempt,
wait_exponential_jitter,
- before_sleep_log,
- RetryError,
)
from repowise.core.providers.llm.base import (
@@ -34,10 +35,10 @@ from repowise.core.providers.llm.base import (
GeneratedResponse,
ProviderError,
RateLimitError,
+ ensure_reasoning_supported,
)
-
-from typing import TYPE_CHECKING, Any, AsyncIterator
from repowise.core.rate_limiter import RateLimiter
+from repowise.core.reasoning import ReasoningMode
if TYPE_CHECKING:
from repowise.core.generation.cost_tracker import CostTracker
@@ -95,7 +96,9 @@ class AnthropicProvider(BaseProvider):
max_tokens: int = 4096,
temperature: float = 0.3,
request_id: str | None = None,
+ reasoning: ReasoningMode = "auto",
) -> GeneratedResponse:
+ ensure_reasoning_supported("anthropic", self._model, reasoning)
if self._rate_limiter:
await self._rate_limiter.acquire(estimated_tokens=max_tokens)
diff --git a/packages/core/src/repowise/core/providers/llm/base.py b/packages/core/src/repowise/core/providers/llm/base.py
index 9e649ad9b..ce8b5ed27 100644
--- a/packages/core/src/repowise/core/providers/llm/base.py
+++ b/packages/core/src/repowise/core/providers/llm/base.py
@@ -17,6 +17,8 @@ from abc import ABC, abstractmethod
from dataclasses import dataclass, field
from typing import Any, AsyncIterator, Protocol, runtime_checkable
+from repowise.core.reasoning import ReasoningMode, normalize_reasoning
+
@dataclass
class GeneratedResponse:
@@ -69,6 +71,7 @@ class BaseProvider(ABC):
max_tokens: int = 4096,
temperature: float = 0.3,
request_id: str | None = None,
+ reasoning: ReasoningMode = "auto",
) -> GeneratedResponse:
"""Generate a response from the LLM.
@@ -81,6 +84,9 @@ class BaseProvider(ABC):
temperature: Sampling temperature. 0.0 is fully deterministic.
repowise uses 0.3 for consistent doc style.
request_id: Optional trace ID for logging and debugging.
+ reasoning: Provider-level reasoning intent. ``auto`` preserves
+ provider defaults; ``off`` and ``minimal`` are
+ translated by providers that support them.
Returns:
GeneratedResponse with content and token usage.
@@ -140,6 +146,29 @@ class RateLimitError(ProviderError):
"""
+def ensure_reasoning_supported(
+ provider: str,
+ model: str,
+ reasoning: ReasoningMode,
+ supported_modes: tuple[ReasoningMode, ...] = (),
+ *,
+ detail: str | None = None,
+) -> ReasoningMode:
+ """Return normalized reasoning mode or fail before issuing an API call."""
+ mode = normalize_reasoning(reasoning)
+ if mode == "auto" or mode in supported_modes:
+ return mode
+
+ supported = ", ".join(dict.fromkeys(("auto", *supported_modes)))
+ message = (
+ f"reasoning={mode!r} is not supported by provider {provider!r} "
+ f"for model {model!r}. Supported reasoning modes: {supported}."
+ )
+ if detail:
+ message = f"{message} {detail}"
+ raise ProviderError(provider, message)
+
+
# ---------------------------------------------------------------------------
# Chat streaming types and protocol (opt-in for providers that support it)
# ---------------------------------------------------------------------------
diff --git a/packages/core/src/repowise/core/providers/llm/deepseek.py b/packages/core/src/repowise/core/providers/llm/deepseek.py
index be9095f45..6864686ab 100644
--- a/packages/core/src/repowise/core/providers/llm/deepseek.py
+++ b/packages/core/src/repowise/core/providers/llm/deepseek.py
@@ -36,8 +36,10 @@ from repowise.core.providers.llm.base import (
GeneratedResponse,
ProviderError,
RateLimitError,
+ ensure_reasoning_supported,
)
from repowise.core.rate_limiter import RateLimiter
+from repowise.core.reasoning import ReasoningMode
if TYPE_CHECKING:
from repowise.core.generation.cost_tracker import CostTracker
@@ -100,7 +102,9 @@ class DeepSeekProvider(BaseProvider):
max_tokens: int = 4096,
temperature: float = 0.3,
request_id: str | None = None,
+ reasoning: ReasoningMode = "auto",
) -> GeneratedResponse:
+ ensure_reasoning_supported("deepseek", self._model, reasoning)
if self._rate_limiter:
await self._rate_limiter.acquire(estimated_tokens=max_tokens)
diff --git a/packages/core/src/repowise/core/providers/llm/gemini.py b/packages/core/src/repowise/core/providers/llm/gemini.py
index 8b9f68a72..12c1c23fc 100644
--- a/packages/core/src/repowise/core/providers/llm/gemini.py
+++ b/packages/core/src/repowise/core/providers/llm/gemini.py
@@ -19,6 +19,8 @@ import structlog
# Suppress "Both GOOGLE_API_KEY and GEMINI_API_KEY are set" from google-genai SDK.
# We resolve and pass the key explicitly, so the env-var conflict warning is noise.
logging.getLogger("google_genai._api_client").setLevel(logging.ERROR)
+from typing import TYPE_CHECKING, Any, AsyncIterator
+
from tenacity import (
RetryError,
retry,
@@ -34,10 +36,10 @@ from repowise.core.providers.llm.base import (
GeneratedResponse,
ProviderError,
RateLimitError,
+ ensure_reasoning_supported,
)
-
-from typing import TYPE_CHECKING, Any, AsyncIterator
from repowise.core.rate_limiter import RateLimiter
+from repowise.core.reasoning import ReasoningMode
if TYPE_CHECKING:
from repowise.core.generation.cost_tracker import CostTracker
@@ -99,7 +101,9 @@ class GeminiProvider(BaseProvider):
max_tokens: int = 4096,
temperature: float = 0.3,
request_id: str | None = None,
+ reasoning: ReasoningMode = "auto",
) -> GeneratedResponse:
+ ensure_reasoning_supported("gemini", self._model, reasoning)
if self._rate_limiter:
await self._rate_limiter.acquire(estimated_tokens=max_tokens)
@@ -392,9 +396,10 @@ def _to_gemini_contents(messages: list[dict[str, Any]]) -> list:
round-trips use native Gemini Content objects (preserving thought
signatures).
"""
- from google.genai import types as genai_types # type: ignore[import-untyped]
import json as _json
+ from google.genai import types as genai_types # type: ignore[import-untyped]
+
contents = []
for msg in messages:
role = msg.get("role", "")
diff --git a/packages/core/src/repowise/core/providers/llm/litellm.py b/packages/core/src/repowise/core/providers/llm/litellm.py
index e73376625..41bf47af5 100644
--- a/packages/core/src/repowise/core/providers/llm/litellm.py
+++ b/packages/core/src/repowise/core/providers/llm/litellm.py
@@ -20,13 +20,15 @@ Reference: https://docs.litellm.ai/docs/providers
from __future__ import annotations
import os
+from typing import TYPE_CHECKING, Any, AsyncIterator
+
import structlog
from tenacity import (
+ RetryError,
retry,
retry_if_exception_type,
stop_after_attempt,
wait_exponential_jitter,
- RetryError,
)
from repowise.core.providers.llm.base import (
@@ -36,10 +38,10 @@ from repowise.core.providers.llm.base import (
GeneratedResponse,
ProviderError,
RateLimitError,
+ ensure_reasoning_supported,
)
-
-from typing import TYPE_CHECKING, Any, AsyncIterator
from repowise.core.rate_limiter import RateLimiter
+from repowise.core.reasoning import ReasoningMode
if TYPE_CHECKING:
from repowise.core.generation.cost_tracker import CostTracker
@@ -98,7 +100,9 @@ class LiteLLMProvider(BaseProvider):
max_tokens: int = 4096,
temperature: float = 0.3,
request_id: str | None = None,
+ reasoning: ReasoningMode = "auto",
) -> GeneratedResponse:
+ ensure_reasoning_supported("litellm", self._model, reasoning)
if self._rate_limiter:
await self._rate_limiter.acquire(estimated_tokens=max_tokens)
@@ -207,6 +211,7 @@ class LiteLLMProvider(BaseProvider):
tool_executor: Any | None = None,
) -> AsyncIterator[ChatStreamEvent]:
import json as _json
+
import litellm # type: ignore[import-untyped]
litellm.set_verbose = False
diff --git a/packages/core/src/repowise/core/providers/llm/mock.py b/packages/core/src/repowise/core/providers/llm/mock.py
index ff23c36d5..247274e76 100644
--- a/packages/core/src/repowise/core/providers/llm/mock.py
+++ b/packages/core/src/repowise/core/providers/llm/mock.py
@@ -32,6 +32,7 @@ import json
from pathlib import Path
from repowise.core.providers.llm.base import BaseProvider, GeneratedResponse
+from repowise.core.reasoning import ReasoningMode
# Resolve fixture directory relative to this file so tests work regardless
# of working directory. Goes up: mock.py → providers/ → core/ → repowise/ →
@@ -108,6 +109,7 @@ class MockProvider(BaseProvider):
max_tokens: int = 4096,
temperature: float = 0.3,
request_id: str | None = None,
+ reasoning: ReasoningMode = "auto",
) -> GeneratedResponse:
"""Return a fixture response without calling any external API."""
self._calls.append(
@@ -117,6 +119,7 @@ class MockProvider(BaseProvider):
"max_tokens": max_tokens,
"temperature": temperature,
"request_id": request_id,
+ "reasoning": reasoning,
}
)
call_idx = self._call_count
diff --git a/packages/core/src/repowise/core/providers/llm/ollama.py b/packages/core/src/repowise/core/providers/llm/ollama.py
index 6b4693c83..0d57f5247 100644
--- a/packages/core/src/repowise/core/providers/llm/ollama.py
+++ b/packages/core/src/repowise/core/providers/llm/ollama.py
@@ -21,15 +21,17 @@ Usage:
from __future__ import annotations
import os
+from typing import Any, AsyncIterator
+
import structlog
-from openai import AsyncOpenAI
from openai import APIStatusError as _OpenAIAPIStatusError
+from openai import AsyncOpenAI
from tenacity import (
+ RetryError,
retry,
retry_if_exception_type,
stop_after_attempt,
wait_exponential_jitter,
- RetryError,
)
from repowise.core.providers.llm.base import (
@@ -38,10 +40,10 @@ from repowise.core.providers.llm.base import (
ChatToolCall,
GeneratedResponse,
ProviderError,
+ ensure_reasoning_supported,
)
-
-from typing import Any, AsyncIterator
from repowise.core.rate_limiter import RateLimiter
+from repowise.core.reasoning import ReasoningMode
log = structlog.get_logger(__name__)
@@ -100,7 +102,9 @@ class OllamaProvider(BaseProvider):
max_tokens: int = 4096,
temperature: float = 0.3,
request_id: str | None = None,
+ reasoning: ReasoningMode = "auto",
) -> GeneratedResponse:
+ ensure_reasoning_supported("ollama", self._model, reasoning)
if self._rate_limiter:
await self._rate_limiter.acquire(estimated_tokens=max_tokens)
diff --git a/packages/core/src/repowise/core/providers/llm/openai.py b/packages/core/src/repowise/core/providers/llm/openai.py
index d7550a098..b40435a1c 100644
--- a/packages/core/src/repowise/core/providers/llm/openai.py
+++ b/packages/core/src/repowise/core/providers/llm/openai.py
@@ -13,17 +13,18 @@ Recommended models (as of 2026):
from __future__ import annotations
import os
+from typing import TYPE_CHECKING, Any, AsyncIterator
import structlog
+from openai import APIStatusError as _OpenAIAPIStatusError
from openai import AsyncOpenAI
from openai import RateLimitError as _OpenAIRateLimitError
-from openai import APIStatusError as _OpenAIAPIStatusError
from tenacity import (
+ RetryError,
retry,
retry_if_exception_type,
stop_after_attempt,
wait_exponential_jitter,
- RetryError,
)
from repowise.core.providers.llm.base import (
@@ -33,10 +34,10 @@ from repowise.core.providers.llm.base import (
GeneratedResponse,
ProviderError,
RateLimitError,
+ ensure_reasoning_supported,
)
-
-from typing import TYPE_CHECKING, Any, AsyncIterator
from repowise.core.rate_limiter import RateLimiter
+from repowise.core.reasoning import ReasoningMode, normalize_reasoning
if TYPE_CHECKING:
from repowise.core.generation.cost_tracker import CostTracker
@@ -46,6 +47,70 @@ log = structlog.get_logger(__name__)
_MAX_RETRIES = 3
_MIN_WAIT = 1.0
_MAX_WAIT = 4.0
+_QWEN_THINKING_MODEL_MARKERS = ("qwen", "qwq")
+_OPENAI_MINIMAL_REASONING_EXACT_MODELS = ("gpt-5", "gpt-5-mini", "gpt-5-nano")
+_OPENAI_MINIMAL_REASONING_PREFIXES = ("gpt-5-mini-", "gpt-5-nano-")
+
+
+def _model_leaf(model: str) -> str:
+ return model.rsplit("/", 1)[-1].lower()
+
+
+def _supports_openai_reasoning_effort(model: str) -> bool:
+ leaf = _model_leaf(model)
+ # Keep this conservative: dotted, pro, codex, and future aliases do not all
+ # accept the same Chat Completions reasoning_effort values.
+ return leaf in _OPENAI_MINIMAL_REASONING_EXACT_MODELS or leaf.startswith(
+ _OPENAI_MINIMAL_REASONING_PREFIXES
+ )
+
+
+def _supports_chat_template_thinking_toggle(model: str) -> bool:
+ leaf = _model_leaf(model)
+ return any(marker in leaf for marker in _QWEN_THINKING_MODEL_MARKERS)
+
+
+def _openai_supported_reasoning_modes(model: str) -> tuple[ReasoningMode, ...]:
+ modes: list[ReasoningMode] = []
+ if _supports_openai_reasoning_effort(model):
+ modes.append("minimal")
+ if _supports_chat_template_thinking_toggle(model):
+ modes.append("off")
+ return tuple(modes)
+
+
+def _resolve_openai_reasoning_mode(
+ reasoning: ReasoningMode, *, model: str
+) -> ReasoningMode:
+ """Validate OpenAI-compatible reasoning support before retry handling."""
+ return ensure_reasoning_supported(
+ "openai",
+ model,
+ normalize_reasoning(reasoning),
+ _openai_supported_reasoning_modes(model),
+ detail=(
+ "OpenAIProvider maps minimal to OpenAI reasoning_effort only for "
+ "known compatible OpenAI model ids (gpt-5, gpt-5-mini, "
+ "gpt-5-nano, and mini/nano snapshot ids), and maps off to "
+ "Qwen/QwQ chat_template_kwargs for OpenAI-compatible endpoints."
+ ),
+ )
+
+
+def _openai_reasoning_kwargs(reasoning: ReasoningMode) -> dict[str, Any]:
+ """Translate a validated repowise reasoning intent to OpenAI kwargs."""
+ mode = normalize_reasoning(reasoning)
+ if mode == "auto":
+ return {}
+ if mode == "minimal":
+ return {"reasoning_effort": "minimal"}
+ return {
+ "extra_body": {
+ "chat_template_kwargs": {
+ "enable_thinking": False,
+ },
+ },
+ }
class OpenAIProvider(BaseProvider):
@@ -93,7 +158,11 @@ class OpenAIProvider(BaseProvider):
max_tokens: int = 4096,
temperature: float = 0.3,
request_id: str | None = None,
+ reasoning: ReasoningMode = "auto",
) -> GeneratedResponse:
+ reasoning_mode = _resolve_openai_reasoning_mode(
+ reasoning, model=self._model
+ )
if self._rate_limiter:
await self._rate_limiter.acquire(estimated_tokens=max_tokens)
@@ -111,6 +180,7 @@ class OpenAIProvider(BaseProvider):
max_tokens=max_tokens,
temperature=temperature,
request_id=request_id,
+ reasoning=reasoning_mode,
)
except RetryError as exc:
raise ProviderError(
@@ -131,16 +201,21 @@ class OpenAIProvider(BaseProvider):
max_tokens: int,
temperature: float,
request_id: str | None,
+ reasoning: ReasoningMode,
) -> GeneratedResponse:
try:
- response = await self._client.chat.completions.create(
- model=self._model,
- max_completion_tokens=max_tokens,
- temperature=temperature,
- messages=[
+ kwargs: dict[str, Any] = {
+ "model": self._model,
+ "max_completion_tokens": max_tokens,
+ "temperature": temperature,
+ "messages": [
{"role": "system", "content": system_prompt},
{"role": "user", "content": user_prompt},
],
+ }
+ kwargs.update(_openai_reasoning_kwargs(reasoning))
+ response = await self._client.chat.completions.create(
+ **kwargs,
)
except _OpenAIRateLimitError as exc:
raise RateLimitError("openai", str(exc), status_code=429) from exc
diff --git a/packages/core/src/repowise/core/providers/llm/openrouter.py b/packages/core/src/repowise/core/providers/llm/openrouter.py
index 85e6391ef..6579fa31a 100644
--- a/packages/core/src/repowise/core/providers/llm/openrouter.py
+++ b/packages/core/src/repowise/core/providers/llm/openrouter.py
@@ -14,17 +14,18 @@ Popular models:
from __future__ import annotations
import os
+from typing import TYPE_CHECKING, Any, AsyncIterator
import structlog
+from openai import APIStatusError as _OpenAIAPIStatusError
from openai import AsyncOpenAI
from openai import RateLimitError as _OpenAIRateLimitError
-from openai import APIStatusError as _OpenAIAPIStatusError
from tenacity import (
+ RetryError,
retry,
retry_if_exception_type,
stop_after_attempt,
wait_exponential_jitter,
- RetryError,
)
from repowise.core.providers.llm.base import (
@@ -34,10 +35,10 @@ from repowise.core.providers.llm.base import (
GeneratedResponse,
ProviderError,
RateLimitError,
+ ensure_reasoning_supported,
)
-
-from typing import TYPE_CHECKING, Any, AsyncIterator
from repowise.core.rate_limiter import RateLimiter
+from repowise.core.reasoning import ReasoningMode, normalize_reasoning
if TYPE_CHECKING:
from repowise.core.generation.cost_tracker import CostTracker
@@ -47,6 +48,62 @@ log = structlog.get_logger(__name__)
_MAX_RETRIES = 3
_MIN_WAIT = 1.0
_MAX_WAIT = 4.0
+_OPENROUTER_OPENAI_REASONING_PREFIXES = ("o1", "o3", "o4")
+_OPENROUTER_OPENAI_GPT5_EXACT_MODELS = ("gpt-5", "gpt-5-mini", "gpt-5-nano")
+_OPENROUTER_OPENAI_GPT5_PREFIXES = ("gpt-5-mini-", "gpt-5-nano-")
+
+
+def _model_leaf(model: str) -> str:
+ return model.rsplit("/", 1)[-1].lower()
+
+
+def _openrouter_supports_reasoning_effort(model: str) -> bool:
+ normalized = model.lower()
+ leaf = _model_leaf(model)
+ if normalized.startswith(("x-ai/grok", "xai/grok")):
+ return True
+ if normalized.startswith("openai/"):
+ if leaf.startswith(_OPENROUTER_OPENAI_REASONING_PREFIXES):
+ return True
+ if leaf in _OPENROUTER_OPENAI_GPT5_EXACT_MODELS:
+ return True
+ if leaf.startswith(_OPENROUTER_OPENAI_GPT5_PREFIXES):
+ return True
+ return False
+
+
+def _resolve_openrouter_reasoning_mode(
+ reasoning: ReasoningMode, *, model: str
+) -> ReasoningMode:
+ """Validate OpenRouter reasoning support before retry handling."""
+ supported_modes: tuple[ReasoningMode, ...] = (
+ ("off", "minimal") if _openrouter_supports_reasoning_effort(model) else ()
+ )
+ return ensure_reasoning_supported(
+ "openrouter",
+ model,
+ normalize_reasoning(reasoning),
+ supported_modes,
+ detail=(
+ "OpenRouter maps reasoning.effort for OpenAI reasoning and Grok "
+ "model families with known effort support. Unknown, dotted, and "
+ "pro GPT-5 routes fail fast until explicitly mapped."
+ ),
+ )
+
+
+def _openrouter_reasoning_kwargs(reasoning: ReasoningMode) -> dict[str, Any]:
+ """Translate a validated repowise reasoning intent to OpenRouter kwargs."""
+ mode = normalize_reasoning(reasoning)
+ if mode == "auto":
+ return {}
+ return {
+ "extra_body": {
+ "reasoning": {
+ "effort": "none" if mode == "off" else "minimal",
+ }
+ }
+ }
class OpenRouterProvider(BaseProvider):
@@ -112,7 +169,11 @@ class OpenRouterProvider(BaseProvider):
max_tokens: int = 4096,
temperature: float = 0.3,
request_id: str | None = None,
+ reasoning: ReasoningMode = "auto",
) -> GeneratedResponse:
+ reasoning_mode = _resolve_openrouter_reasoning_mode(
+ reasoning, model=self._model
+ )
if self._rate_limiter:
await self._rate_limiter.acquire(estimated_tokens=max_tokens)
@@ -130,6 +191,7 @@ class OpenRouterProvider(BaseProvider):
max_tokens=max_tokens,
temperature=temperature,
request_id=request_id,
+ reasoning=reasoning_mode,
)
except RetryError as exc:
raise ProviderError(
@@ -150,17 +212,20 @@ class OpenRouterProvider(BaseProvider):
max_tokens: int,
temperature: float,
request_id: str | None,
+ reasoning: ReasoningMode,
) -> GeneratedResponse:
try:
- response = await self._client.chat.completions.create(
- model=self._model,
- max_tokens=max_tokens,
- temperature=temperature,
- messages=[
+ kwargs: dict[str, Any] = {
+ "model": self._model,
+ "max_tokens": max_tokens,
+ "temperature": temperature,
+ "messages": [
{"role": "system", "content": system_prompt},
{"role": "user", "content": user_prompt},
],
- )
+ }
+ kwargs.update(_openrouter_reasoning_kwargs(reasoning))
+ response = await self._client.chat.completions.create(**kwargs)
except _OpenAIRateLimitError as exc:
raise RateLimitError("openrouter", str(exc), status_code=429) from exc
except _OpenAIAPIStatusError as exc:
diff --git a/packages/core/src/repowise/core/reasoning.py b/packages/core/src/repowise/core/reasoning.py
new file mode 100644
index 000000000..c4cf5213b
--- /dev/null
+++ b/packages/core/src/repowise/core/reasoning.py
@@ -0,0 +1,40 @@
+"""Shared reasoning-mode helpers for generation providers."""
+
+from __future__ import annotations
+
+import os
+from typing import Literal, cast
+
+ReasoningMode = Literal["auto", "off", "minimal"]
+REASONING_MODES: tuple[ReasoningMode, ...] = ("auto", "off", "minimal")
+
+
+def normalize_reasoning(
+ value: object,
+ *,
+ default: ReasoningMode = "auto",
+) -> ReasoningMode:
+ """Normalize a user/config supplied reasoning mode.
+
+ ``auto`` preserves provider defaults. ``off`` and ``minimal`` are provider
+ intents; each provider decides whether and how to translate them.
+ """
+ if value is None:
+ return default
+
+ normalized = str(value).strip().lower()
+ if normalized not in REASONING_MODES:
+ choices = ", ".join(REASONING_MODES)
+ raise ValueError(f"reasoning must be one of: {choices}")
+ return cast(ReasoningMode, normalized)
+
+
+def resolve_reasoning(
+ reasoning: str | None = None,
+ config: dict[str, object] | None = None,
+) -> ReasoningMode:
+ """Resolve reasoning from explicit value, env, repo config, then default."""
+ raw = reasoning or os.environ.get("REPOWISE_REASONING")
+ if raw is None and config:
+ raw = config.get("reasoning")
+ return normalize_reasoning(raw)
diff --git a/packages/core/src/repowise/core/repo_config.py b/packages/core/src/repowise/core/repo_config.py
new file mode 100644
index 000000000..bfb1a67eb
--- /dev/null
+++ b/packages/core/src/repowise/core/repo_config.py
@@ -0,0 +1,48 @@
+"""Repo-local configuration helpers shared by CLI, server, and core paths."""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import Any
+
+CONFIG_FILENAME = "config.yaml"
+
+
+def get_repowise_dir(repo_path: Path | str) -> Path:
+ """Return the repo-local ``.repowise`` directory."""
+ return Path(repo_path) / ".repowise"
+
+
+def load_repo_config(repo_path: Path | str) -> dict[str, Any]:
+ """Load ``.repowise/config.yaml`` or return an empty dict if absent."""
+ config_path = get_repowise_dir(repo_path) / CONFIG_FILENAME
+ if not config_path.exists():
+ return {}
+
+ text = config_path.read_text(encoding="utf-8")
+ try:
+ import yaml # type: ignore[import-untyped]
+
+ result = yaml.safe_load(text) or {}
+ if isinstance(result, dict) and isinstance(result.get("reasoning"), bool):
+ raw_reasoning = _read_flat_scalar(text, "reasoning")
+ if raw_reasoning:
+ result["reasoning"] = raw_reasoning
+ return result
+ except ImportError:
+ # Simple line-by-line parser for the flat key: value format we write.
+ result: dict[str, Any] = {}
+ for line in text.splitlines():
+ if ":" in line:
+ key, _, value = line.partition(":")
+ result[key.strip()] = value.strip()
+ return result
+
+
+def _read_flat_scalar(text: str, key: str) -> str | None:
+ """Read a top-level scalar from config text before YAML bool coercion."""
+ for line in text.splitlines():
+ current_key, separator, value = line.partition(":")
+ if separator and current_key.strip() == key:
+ return value.split("#", 1)[0].strip().strip("'\"")
+ return None
diff --git a/packages/server/src/repowise/server/job_executor.py b/packages/server/src/repowise/server/job_executor.py
index 305005653..9add0e26a 100644
--- a/packages/server/src/repowise/server/job_executor.py
+++ b/packages/server/src/repowise/server/job_executor.py
@@ -461,10 +461,14 @@ async def _incremental_page_regen(
affected_source = {p: s for p, s in result.source_map.items() if p in regen_set}
from repowise.core.generation import ContextAssembler, GenerationConfig, PageGenerator
+ from repowise.core.reasoning import resolve_reasoning
+ from repowise.core.repo_config import load_repo_config
- config = GenerationConfig()
- assembler = ContextAssembler(config)
- generator = PageGenerator(llm_client, assembler, config)
+ generation_config = GenerationConfig(
+ reasoning=resolve_reasoning(config=load_repo_config(repo_path))
+ )
+ assembler = ContextAssembler(generation_config)
+ generator = PageGenerator(llm_client, assembler, generation_config)
pages = await generator.generate_all(
affected_parsed,
diff --git a/tests/unit/cli/test_commands.py b/tests/unit/cli/test_commands.py
index b249adbbe..0da4bab5b 100644
--- a/tests/unit/cli/test_commands.py
+++ b/tests/unit/cli/test_commands.py
@@ -37,11 +37,13 @@ class TestCliBasics:
assert "--provider" in result.output
assert "--dry-run" in result.output
assert "--skip-tests" in result.output
+ assert "--reasoning" in result.output
def test_update_help(self, runner):
result = runner.invoke(cli, ["update", "--help"])
assert result.exit_code == 0
assert "--since" in result.output
+ assert "--reasoning" in result.output
def test_search_help(self, runner):
result = runner.invoke(cli, ["search", "--help"])
diff --git a/tests/unit/cli/test_helpers.py b/tests/unit/cli/test_helpers.py
index 7b5984492..0383b5da9 100644
--- a/tests/unit/cli/test_helpers.py
+++ b/tests/unit/cli/test_helpers.py
@@ -8,13 +8,14 @@ from typing import Any
import pytest
from repowise.cli.helpers import (
- ensure_repowise_dir,
CONFIG_FILENAME,
+ ensure_repowise_dir,
get_db_url_for_repo,
get_head_commit,
get_repowise_dir,
load_state,
resolve_provider,
+ resolve_reasoning,
resolve_repo_path,
run_async,
save_state,
@@ -100,6 +101,20 @@ class TestDbUrl:
assert (tmp_path / ".repowise").exists()
+class TestResolveReasoning:
+ def test_flag_wins_over_env(self, monkeypatch):
+ monkeypatch.setenv("REPOWISE_REASONING", "minimal")
+ assert resolve_reasoning("off", {"reasoning": "auto"}) == "off"
+
+ def test_env_wins_over_config(self, monkeypatch):
+ monkeypatch.setenv("REPOWISE_REASONING", "minimal")
+ assert resolve_reasoning(config={"reasoning": "off"}) == "minimal"
+
+ def test_config_wins_over_default(self, monkeypatch):
+ monkeypatch.delenv("REPOWISE_REASONING", raising=False)
+ assert resolve_reasoning(config={"reasoning": "off"}) == "off"
+
+
# ---------------------------------------------------------------------------
# State file
# ---------------------------------------------------------------------------
diff --git a/tests/unit/generation/test_models.py b/tests/unit/generation/test_models.py
index 2096bb511..fab450a4d 100644
--- a/tests/unit/generation/test_models.py
+++ b/tests/unit/generation/test_models.py
@@ -4,6 +4,8 @@ from __future__ import annotations
from datetime import UTC, datetime, timedelta
+import pytest
+
from repowise.core.generation.models import (
ConfidenceDecayResult,
GeneratedPage,
@@ -152,8 +154,19 @@ def test_generation_config_defaults():
assert config.expiry_threshold_days == 30
assert config.top_symbol_percentile == 0.10
assert config.large_file_source_pct == 0.4
+ assert config.reasoning == "auto"
def test_generation_config_embed_concurrency_defaults_to_max_concurrency():
config = GenerationConfig(max_concurrency=3)
assert config.embed_concurrency == 3
+
+
+def test_generation_config_normalizes_reasoning():
+ config = GenerationConfig(reasoning="OFF")
+ assert config.reasoning == "off"
+
+
+def test_generation_config_rejects_invalid_reasoning():
+ with pytest.raises(ValueError):
+ GenerationConfig(reasoning="verbose")
diff --git a/tests/unit/generation/test_page_generator.py b/tests/unit/generation/test_page_generator.py
index a035fcb0b..8df5f2cb7 100644
--- a/tests/unit/generation/test_page_generator.py
+++ b/tests/unit/generation/test_page_generator.py
@@ -108,6 +108,26 @@ async def test_generate_file_page_increments_call_count(
assert provider.call_count == 1
+async def test_generate_file_page_forwards_reasoning_config(
+ sample_parsed_file, sample_graph, graph_metrics, sample_source_bytes
+):
+ provider = MockProvider()
+ config = GenerationConfig(reasoning="off")
+ assembler = ContextAssembler(config)
+ gen = PageGenerator(provider, assembler, config)
+
+ await gen.generate_file_page(
+ sample_parsed_file,
+ sample_graph,
+ graph_metrics["pagerank"],
+ graph_metrics["betweenness"],
+ graph_metrics["community"],
+ sample_source_bytes,
+ )
+
+ assert provider.calls[0]["reasoning"] == "off"
+
+
# ---------------------------------------------------------------------------
# Prompt cache
# ---------------------------------------------------------------------------
diff --git a/tests/unit/generation/test_pipeline_reasoning.py b/tests/unit/generation/test_pipeline_reasoning.py
new file mode 100644
index 000000000..d7b83eb0c
--- /dev/null
+++ b/tests/unit/generation/test_pipeline_reasoning.py
@@ -0,0 +1,33 @@
+"""Reasoning propagation tests for programmatic generation paths."""
+
+from __future__ import annotations
+
+from repowise.core.pipeline import run_pipeline
+from repowise.core.providers.llm.mock import MockProvider
+
+
+async def test_run_pipeline_loads_repo_reasoning_config(tmp_path):
+ repo_path = tmp_path / "repo"
+ repo_path.mkdir()
+ (repo_path / "main.py").write_text(
+ 'def hello() -> str:\n return "hello"\n',
+ encoding="utf-8",
+ )
+ (repo_path / ".repowise").mkdir()
+ (repo_path / ".repowise" / "config.yaml").write_text(
+ "reasoning: off # disable thinking for Qwen3-style models\n",
+ encoding="utf-8",
+ )
+
+ provider = MockProvider()
+ result = await run_pipeline(
+ repo_path,
+ generate_docs=True,
+ llm_client=provider,
+ concurrency=1,
+ test_run=True,
+ )
+
+ assert result.generated_pages
+ assert provider.calls
+ assert all(call["reasoning"] == "off" for call in provider.calls)
diff --git a/tests/unit/test_providers/test_deepseek_provider.py b/tests/unit/test_providers/test_deepseek_provider.py
index f46748360..a265f67f3 100644
--- a/tests/unit/test_providers/test_deepseek_provider.py
+++ b/tests/unit/test_providers/test_deepseek_provider.py
@@ -11,7 +11,11 @@ import pytest
pytest.importorskip("openai", reason="openai SDK not installed")
-from repowise.core.providers.llm.base import ChatStreamEvent, GeneratedResponse, ProviderError, RateLimitError
+from repowise.core.providers.llm.base import (
+ GeneratedResponse,
+ ProviderError,
+ RateLimitError,
+)
from repowise.core.providers.llm.deepseek import DeepSeekProvider
@@ -127,6 +131,17 @@ async def test_generate_uses_correct_model_name():
assert kwargs["model"] == "deepseek-v4-flash"
+async def test_generate_rejects_unsupported_reasoning_mode():
+ provider = DeepSeekProvider(api_key="sk-test")
+
+ with patch("openai.AsyncOpenAI") as mock_client:
+ provider._client = mock_client.return_value
+ with pytest.raises(ProviderError, match="reasoning='off' is not supported"):
+ await provider.generate("system", "user", reasoning="off")
+
+ mock_client.return_value.chat.completions.create.assert_not_called()
+
+
async def test_generate_rate_limit_retry():
from openai import RateLimitError as _OpenAIRateLimitError
diff --git a/tests/unit/test_providers/test_openai_provider.py b/tests/unit/test_providers/test_openai_provider.py
index 28466310b..3cf2b49a0 100644
--- a/tests/unit/test_providers/test_openai_provider.py
+++ b/tests/unit/test_providers/test_openai_provider.py
@@ -116,11 +116,86 @@ async def test_generate_sends_correct_messages():
assert kw["model"] == "gpt-5.4-mini"
assert kw["max_completion_tokens"] == 2048
assert kw["temperature"] == 0.5
+ assert "reasoning_effort" not in kw
+ assert "extra_body" not in kw
messages = kw["messages"]
assert messages[0] == {"role": "system", "content": "system msg"}
assert messages[1] == {"role": "user", "content": "user msg"}
+async def test_generate_forwards_minimal_reasoning_effort():
+ provider = OpenAIProvider(api_key="sk-test", model="gpt-5-mini")
+ mock_response = _make_mock_chat_response()
+ captured_kwargs: list[dict] = []
+
+ async def fake_create(**kwargs):
+ captured_kwargs.append(kwargs)
+ return mock_response
+
+ with patch("openai.AsyncOpenAI") as mock_client:
+ mock_client.return_value.chat.completions.create = fake_create
+ provider._client = mock_client.return_value
+ await provider.generate("system msg", "user msg", reasoning="minimal")
+
+ assert captured_kwargs[0]["reasoning_effort"] == "minimal"
+
+
+async def test_generate_forwards_off_reasoning_extra_body():
+ provider = OpenAIProvider(api_key="sk-test", model="qwen3")
+ mock_response = _make_mock_chat_response()
+ captured_kwargs: list[dict] = []
+
+ async def fake_create(**kwargs):
+ captured_kwargs.append(kwargs)
+ return mock_response
+
+ with patch("openai.AsyncOpenAI") as mock_client:
+ mock_client.return_value.chat.completions.create = fake_create
+ provider._client = mock_client.return_value
+ await provider.generate("system msg", "user msg", reasoning="off")
+
+ assert captured_kwargs[0]["extra_body"] == {
+ "chat_template_kwargs": {"enable_thinking": False}
+ }
+
+
+async def test_generate_rejects_minimal_for_non_reasoning_model():
+ provider = OpenAIProvider(api_key="sk-test", model="gpt-4o")
+
+ with patch("openai.AsyncOpenAI") as mock_client:
+ provider._client = mock_client.return_value
+ with pytest.raises(ProviderError, match="reasoning='minimal' is not supported"):
+ await provider.generate("system msg", "user msg", reasoning="minimal")
+
+ mock_client.return_value.chat.completions.create.assert_not_called()
+
+
+ "model",
+ ["gpt-5.1", "gpt-5.2", "gpt-5-pro", "gpt-5-codex", "gpt-5.4-nano"],
+)
+async def test_generate_rejects_minimal_for_known_unsupported_reasoning_models(model):
+ provider = OpenAIProvider(api_key="sk-test", model=model)
+
+ with patch("openai.AsyncOpenAI") as mock_client:
+ provider._client = mock_client.return_value
+ with pytest.raises(ProviderError, match="reasoning='minimal' is not supported"):
+ await provider.generate("system msg", "user msg", reasoning="minimal")
+
+ mock_client.return_value.chat.completions.create.assert_not_called()
+
+
+async def test_generate_rejects_off_for_non_qwen_model():
+ provider = OpenAIProvider(api_key="sk-test", model="gpt-5-mini")
+
+ with patch("openai.AsyncOpenAI") as mock_client:
+ provider._client = mock_client.return_value
+ with pytest.raises(ProviderError, match="reasoning='off' is not supported"):
+ await provider.generate("system msg", "user msg", reasoning="off")
+
+ mock_client.return_value.chat.completions.create.assert_not_called()
+
+
# ---------------------------------------------------------------------------
# Error mapping
# ---------------------------------------------------------------------------
diff --git a/tests/unit/test_providers/test_openrouter_provider.py b/tests/unit/test_providers/test_openrouter_provider.py
index 4fce0d7d9..d8f85ad6c 100644
--- a/tests/unit/test_providers/test_openrouter_provider.py
+++ b/tests/unit/test_providers/test_openrouter_provider.py
@@ -149,11 +149,76 @@ async def test_generate_sends_correct_messages():
assert kw["model"] == "google/gemini-3.1-flash-lite-preview"
assert kw["max_tokens"] == 2048
assert kw["temperature"] == 0.5
+ assert "extra_body" not in kw
messages = kw["messages"]
assert messages[0] == {"role": "system", "content": "system msg"}
assert messages[1] == {"role": "user", "content": "user msg"}
+async def test_generate_forwards_minimal_reasoning_extra_body():
+ provider = OpenRouterProvider(api_key="sk-or-test", model="openai/gpt-5")
+ mock_response = _make_mock_chat_response()
+ captured_kwargs: list[dict] = []
+
+ async def fake_create(**kwargs):
+ captured_kwargs.append(kwargs)
+ return mock_response
+
+ with patch("openai.AsyncOpenAI") as mock_client:
+ mock_client.return_value.chat.completions.create = fake_create
+ provider._client = mock_client.return_value
+ await provider.generate("system msg", "user msg", reasoning="minimal")
+
+ assert captured_kwargs[0]["extra_body"] == {
+ "reasoning": {"effort": "minimal"}
+ }
+
+
+async def test_generate_forwards_off_reasoning_extra_body():
+ provider = OpenRouterProvider(api_key="sk-or-test", model="x-ai/grok-4")
+ mock_response = _make_mock_chat_response()
+ captured_kwargs: list[dict] = []
+
+ async def fake_create(**kwargs):
+ captured_kwargs.append(kwargs)
+ return mock_response
+
+ with patch("openai.AsyncOpenAI") as mock_client:
+ mock_client.return_value.chat.completions.create = fake_create
+ provider._client = mock_client.return_value
+ await provider.generate("system msg", "user msg", reasoning="off")
+
+ assert captured_kwargs[0]["extra_body"] == {"reasoning": {"effort": "none"}}
+
+
+async def test_generate_rejects_reasoning_for_unsupported_openrouter_model():
+ provider = OpenRouterProvider(
+ api_key="sk-or-test", model="google/gemini-3.1-flash-lite-preview"
+ )
+
+ with patch("openai.AsyncOpenAI") as mock_client:
+ provider._client = mock_client.return_value
+ with pytest.raises(ProviderError, match="reasoning='minimal' is not supported"):
+ await provider.generate("system msg", "user msg", reasoning="minimal")
+
+ mock_client.return_value.chat.completions.create.assert_not_called()
+
+
+ "model",
+ ["openai/gpt-5.2", "openai/gpt-5-pro", "openai/gpt-5-codex", "openai/gpt-5.4-nano"],
+)
+async def test_generate_rejects_reasoning_for_known_unsupported_openai_route(model):
+ provider = OpenRouterProvider(api_key="sk-or-test", model=model)
+
+ with patch("openai.AsyncOpenAI") as mock_client:
+ provider._client = mock_client.return_value
+ with pytest.raises(ProviderError, match="reasoning='minimal' is not supported"):
+ await provider.generate("system msg", "user msg", reasoning="minimal")
+
+ mock_client.return_value.chat.completions.create.assert_not_called()
+
+
# ---------------------------------------------------------------------------
# Error mapping
# ---------------------------------------------------------------------------
diff --git a/website/cli-reference.md b/website/cli-reference.md
index 29ba7f8ae..d0906e12c 100644
--- a/website/cli-reference.md
+++ b/website/cli-reference.md
@@ -44,7 +44,7 @@ repowise init [PATH] [OPTIONS]
| Flag | Type | Default | Description |
|------|------|---------|-------------|
-| `--provider` | string | auto | LLM provider: `anthropic`, `openai`, `gemini`, `ollama`, `litellm`, `mock` |
+| `--provider` | string | auto | LLM provider: `anthropic`, `openai`, `openrouter`, `gemini`, `deepseek`, `ollama`, `litellm`, `mock` |
| `--model` | string | — | Model override (e.g., `claude-sonnet-4-6`, `gpt-4.1`) |
| `--embedder` | choice | auto | Embedding provider: `gemini`, `openai`, `mock` |
| `--index-only` | flag | false | Skip LLM generation — parse, graph, git, dead code only |
@@ -55,6 +55,7 @@ repowise init [PATH] [OPTIONS]
| `--exclude` / `-x` | string | — | Gitignore-style exclude pattern (repeatable: `-x vendor/ -x "*.gen.*"`) |
| `--include-submodules` | flag | false | Include git submodule directories (excluded by default) |
| `--concurrency` | int | 5 | Max concurrent LLM calls |
+| `--reasoning` | choice | auto | Reasoning mode for supported providers: `auto`, `off`, or `minimal` |
| `--resume` | flag | false | Resume from last checkpoint after an interruption |
| `--force` | flag | false | Regenerate all pages even if up to date |
| `--commit-limit` | int | 500 | Max commits per file for git analysis (max 5000, saved to config) |
@@ -71,6 +72,12 @@ repowise init
# Non-interactive, specific provider
repowise init --provider anthropic --yes
+# OpenAI-compatible Qwen3 endpoint with thinking disabled
+repowise init --provider openai --model qwen3 --reasoning off
+
+# OpenRouter with minimal reasoning effort
+repowise init --provider openrouter --model openai/gpt-5 --reasoning minimal
+
# Analysis only — free, no API key needed
repowise init --index-only
@@ -121,6 +128,7 @@ repowise update [PATH] [OPTIONS]
| `--provider` | string | — | Override LLM provider |
| `--model` | string | — | Override model |
| `--since` | string | — | Git ref to diff from (overrides saved `state.json`) |
+| `--reasoning` | choice | auto | Reasoning mode for supported providers: `auto`, `off`, or `minimal` |
| `--cascade-budget` | int | auto | Max pages to regenerate from cascading changes |
| `--dry-run` | flag | false | Show affected pages without regenerating |
@@ -129,6 +137,7 @@ repowise update [PATH] [OPTIONS]
```bash
repowise update # Sync since last indexed commit
repowise update --since HEAD~10 # Re-sync last 10 commits
+repowise update --reasoning off # One-off supported-provider thinking-off run
repowise update --dry-run # Preview what would change
repowise update --cascade-budget 20 # Limit cascade regeneration
```
diff --git a/website/configuration.md b/website/configuration.md
index e3edcefa3..dededdd7c 100644
--- a/website/configuration.md
+++ b/website/configuration.md
@@ -46,6 +46,7 @@ The main configuration file. Created after first `init`, updated when you pass `
provider: anthropic # LLM provider
model: claude-sonnet-4-6 # Model identifier
embedder: gemini # Embedding provider
+reasoning: auto # auto | off | minimal
exclude_patterns: # Gitignore-style patterns
- vendor/
- "*.generated.*"
@@ -56,6 +57,18 @@ follow_renames: false # Track file renames in git history
You can edit this file directly. Changes take effect on the next `init`, `update`, or `serve` run.
+`reasoning` controls documentation-generation calls for reasoning-capable chat
+models. `auto` preserves provider defaults. `off` disables Qwen3-style thinking
+for OpenAI-compatible vLLM/SGLang endpoints by sending
+`extra_body.chat_template_kwargs.enable_thinking=false`, and maps to
+OpenRouter's `reasoning.effort=none` for effort-capable OpenRouter model
+families. `minimal` requests OpenAI's lowest supported reasoning effort for
+supported OpenAI reasoning models and maps to OpenRouter's
+`reasoning.effort=minimal` for effort-capable OpenRouter model families.
+Providers or models that cannot translate an explicit mode fail before making
+an API call. The interactive `serve` chat streaming path is separate and does
+not currently consume this setting.
+
---
## LLM providers
@@ -93,6 +106,20 @@ export OPENAI_API_KEY="sk-..."
repowise init --provider openai --model gpt-4.1
```
+For an OpenAI-compatible Qwen3 endpoint served by vLLM or SGLang:
+
+```bash
+export OPENAI_BASE_URL="http://localhost:8000/v1"
+repowise init --provider openai --model qwen3 --reasoning off
+```
+
+For OpenRouter:
+
+```bash
+repowise init --provider openrouter --model openai/gpt-5 --reasoning minimal
+repowise init --provider openrouter --model x-ai/grok-4 --reasoning off
+```
+
---
### Gemini (Google)