Skip to content

Your lightweight open-source personal agent.

refactor(agent): remove context block budget override (#5668)

chengyongruf37ba60ed489
diff --git a/nanobot/agent/context_governance.py b/nanobot/agent/context_governance.py
index fef4c4b5c..e552ff133 100644
--- a/nanobot/agent/context_governance.py
+++ b/nanobot/agent/context_governance.py
@@ -122,7 +122,6 @@ class ContextGovernanceConfig:
session_key: str | None
max_tool_result_chars: int
context_window_tokens: int | None = None
- context_block_limit: int | None = None
max_tokens: int | None = None
@@ -699,9 +698,7 @@ class ContextGovernor:
max_output = config.max_tokens if isinstance(config.max_tokens, int) else (
provider_max_tokens if isinstance(provider_max_tokens, int) else 4096
)
- budget = config.context_block_limit or (
- config.context_window_tokens - max_output - SNIP_SAFETY_BUFFER
- )
+ budget = config.context_window_tokens - max_output - SNIP_SAFETY_BUFFER
return budget if budget > 0 else 0
@staticmethod
diff --git a/nanobot/agent/loop.py b/nanobot/agent/loop.py
index 14846f0be..583bd3d83 100644
--- a/nanobot/agent/loop.py
+++ b/nanobot/agent/loop.py
@@ -274,7 +274,6 @@ class AgentLoop:
max_iterations: int | None = None,
max_concurrent_subagents: int | None = None,
context_window_tokens: int | None = None,
- context_block_limit: int | None = None,
max_tool_result_chars: int | None = None,
provider_retry_mode: str = "standard",
tool_hint_max_length: int | None = None,
@@ -355,7 +354,6 @@ class AgentLoop:
preset_snapshot_loader=preset_snapshot_loader,
)
self.dream_model_preset = dream_model_preset
- self.context_block_limit = context_block_limit
self.max_tool_result_chars = (
max_tool_result_chars
if max_tool_result_chars is not None
@@ -516,7 +514,6 @@ class AgentLoop:
max_iterations=defaults.max_tool_iterations,
max_concurrent_subagents=defaults.max_concurrent_subagents,
context_window_tokens=context_window_tokens,
- context_block_limit=defaults.context_block_limit,
max_tool_result_chars=defaults.max_tool_result_chars,
provider_retry_mode=defaults.provider_retry_mode,
tool_hint_max_length=defaults.tool_hint_max_length,
@@ -1201,7 +1198,6 @@ class AgentLoop:
concurrent_tools=True,
workspace=effective_scope.project_path,
session_key=session.key if session else None,
- context_block_limit=self.context_block_limit,
provider_retry_mode=self.provider_retry_mode,
retry_wait_callback=on_retry_wait,
checkpoint_callback=_checkpoint,
diff --git a/nanobot/agent/runner.py b/nanobot/agent/runner.py
index 69aed146e..689b8d90b 100644
--- a/nanobot/agent/runner.py
+++ b/nanobot/agent/runner.py
@@ -104,7 +104,6 @@ class AgentRunSpec:
concurrent_tools: bool = False
workspace: Path | None = None
session_key: str | None = None
- context_block_limit: int | None = None
provider_retry_mode: str = "standard"
retry_wait_callback: RetryWaitCallback | None = None
checkpoint_callback: CheckpointCallback | None = None
@@ -426,7 +425,6 @@ class AgentRunner:
session_key=spec.session_key,
max_tool_result_chars=spec.max_tool_result_chars,
context_window_tokens=spec.runtime.context_window_tokens,
- context_block_limit=spec.context_block_limit,
max_tokens=spec.runtime.generation.max_tokens,
)
request_state = ModelRequestState(
diff --git a/nanobot/config/schema.py b/nanobot/config/schema.py
index 8438352f4..c3e75dcaf 100644
--- a/nanobot/config/schema.py
+++ b/nanobot/config/schema.py
@@ -124,7 +124,6 @@ class AgentDefaults(Base):
)
max_tokens: int = 8192
context_window_tokens: int = 200_000
- context_block_limit: int | None = None
temperature: float = 0.1
fallback_models: list[FallbackCandidate] = Field(default_factory=list)
max_tool_iterations: int = 200
diff --git a/tests/agent/test_loop_consolidation_tokens.py b/tests/agent/test_loop_consolidation_tokens.py
index f443e106b..e950e9b40 100644
--- a/tests/agent/test_loop_consolidation_tokens.py
+++ b/tests/agent/test_loop_consolidation_tokens.py
@@ -33,9 +33,6 @@ def _make_loop(
workspace=tmp_path,
model="test-model",
context_window_tokens=context_window_tokens,
- # These tests isolate Memory consolidation; Runner request fitting is
- # covered separately with realistic context windows.
- context_block_limit=10_000,
)
loop.tools.get_definitions = MagicMock(return_value=[])
loop.consolidator._SAFETY_BUFFER = 0
@@ -44,9 +41,9 @@ def _make_loop(
@pytest.mark.asyncio
async def test_runner_pressure_commits_summary_and_current_delta(tmp_path) -> None:
- loop = _make_loop(tmp_path, estimated_tokens=100, context_window_tokens=2_000)
- loop.context_block_limit = 500
- loop.provider.generation = GenerationSettings(max_tokens=100)
+ loop = _make_loop(
+ tmp_path, estimated_tokens=100, context_window_tokens=1_624, max_tokens=100,
+ )
loop.provider.can_resume_conversation_state.return_value = False
loop.schedule_background = lambda coro: coro.close() # type: ignore[method-assign]
diff --git a/tests/agent/test_runner_core.py b/tests/agent/test_runner_core.py
index 1c4be1fa1..b68448115 100644
--- a/tests/agent/test_runner_core.py
+++ b/tests/agent/test_runner_core.py
@@ -423,8 +423,7 @@ async def test_runner_preserves_tool_result_before_rejecting_unfit_followup():
],
tools=tools,
model="gpt-5.6",
- context_window_tokens=3_000,
- context_block_limit=200,
+ context_window_tokens=2_224,
max_tokens=1_000,
max_iterations=3,
max_tool_result_chars=10_000,
diff --git a/tests/agent/test_runner_governance.py b/tests/agent/test_runner_governance.py
index ad292dc6f..5f9c32f22 100644
--- a/tests/agent/test_runner_governance.py
+++ b/tests/agent/test_runner_governance.py
@@ -56,7 +56,6 @@ def _governance_config(
session_key=spec.session_key,
max_tool_result_chars=spec.max_tool_result_chars,
context_window_tokens=spec.runtime.context_window_tokens,
- context_block_limit=spec.context_block_limit,
max_tokens=spec.runtime.generation.max_tokens,
)
@@ -136,8 +135,7 @@ async def test_runner_locally_fits_oversized_initial_transcript(monkeypatch):
],
tools=tools,
model="local-model",
- context_window_tokens=2_000,
- context_block_limit=500,
+ context_window_tokens=1_624,
max_tokens=100,
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
@@ -210,8 +208,7 @@ async def test_runner_summarizes_history_and_preserves_current_input(monkeypatch
provider_state=prior_state,
tools=tools,
model="test-model",
- context_window_tokens=2_000,
- context_block_limit=500,
+ context_window_tokens=1_624,
max_tokens=100,
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
@@ -261,8 +258,7 @@ async def test_runner_rejects_oversized_delta_without_summarizable_history(monke
consolidate_history=consolidate,
tools=tools,
model="test-model",
- context_window_tokens=2_000,
- context_block_limit=500,
+ context_window_tokens=1_624,
max_tokens=100,
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
@@ -313,8 +309,7 @@ async def test_runner_governs_history_before_summarizing_it(monkeypatch):
consolidate_history=consolidate,
tools=tools,
model="test-model",
- context_window_tokens=2_000,
- context_block_limit=500,
+ context_window_tokens=1_624,
max_tokens=100,
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
@@ -395,8 +390,7 @@ async def test_native_compaction_uses_provider_request_boundary(
consolidate_provider_compaction=consolidate_native,
tools=tools,
model="test-model",
- context_window_tokens=2_000,
- context_block_limit=500,
+ context_window_tokens=1_624,
max_tokens=100,
max_iterations=2,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
@@ -465,8 +459,7 @@ async def test_runner_keeps_current_tool_exchange_outside_summary(monkeypatch):
consolidate_history=consolidate,
tools=tools,
model="test-model",
- context_window_tokens=2_000,
- context_block_limit=500,
+ context_window_tokens=1_624,
max_tokens=100,
max_iterations=2,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
@@ -537,8 +530,7 @@ async def test_repeated_pressure_advances_summary_boundary(monkeypatch):
consolidate_history=consolidate,
tools=tools,
model="test-model",
- context_window_tokens=2_000,
- context_block_limit=500,
+ context_window_tokens=1_624,
max_tokens=100,
max_iterations=3,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
@@ -596,8 +588,7 @@ async def test_runner_refuses_checkpoint_that_cannot_fit_with_delta(monkeypatch)
consolidate_history=consolidate,
tools=tools,
model="test-model",
- context_window_tokens=2_000,
- context_block_limit=500,
+ context_window_tokens=1_624,
max_tokens=100,
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
@@ -638,8 +629,8 @@ async def test_runner_governs_messages_added_by_before_iteration_hook(monkeypatc
initial_messages=[{"role": "user", "content": "hello"}],
tools=tools,
model="local-model",
- context_window_tokens=2_000,
- context_block_limit=500,
+ context_window_tokens=1_624,
+ max_tokens=100,
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
hook=MutatingHook(),
@@ -703,8 +694,8 @@ async def test_runner_drops_resumable_provider_state_when_request_is_fitted(monk
],
tools=tools,
model="local-model",
- context_window_tokens=2_000,
- context_block_limit=500,
+ context_window_tokens=1_624,
+ max_tokens=100,
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
provider_state=saved_state,
@@ -760,8 +751,8 @@ async def test_runner_fits_each_malformed_retry_with_its_actual_tools(monkeypatc
initial_messages=[{"role": "user", "content": "use a tool"}],
tools=tools,
model="local-model",
- context_window_tokens=2_000,
- context_block_limit=500,
+ context_window_tokens=1_624,
+ max_tokens=100,
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
))
@@ -825,8 +816,8 @@ async def test_runner_fits_empty_response_finalization_before_dispatch(monkeypat
initial_messages=[{"role": "user", "content": "do task"}],
tools=tools,
model="local-model",
- context_window_tokens=2_000,
- context_block_limit=500,
+ context_window_tokens=1_624,
+ max_tokens=100,
max_iterations=3,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
))
@@ -883,8 +874,8 @@ async def test_runner_fits_max_iteration_finalization_before_dispatch(monkeypatc
initial_messages=[{"role": "user", "content": "inspect"}],
tools=tools,
model="local-model",
- context_window_tokens=2_000,
- context_block_limit=500,
+ context_window_tokens=1_624,
+ max_tokens=100,
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
))
@@ -916,8 +907,8 @@ async def test_matching_reported_provider_usage_avoids_local_estimate(
initial_messages=[{"role": "user", "content": "hello"}],
tools=tools,
model="local-model",
- context_window_tokens=2_000,
- context_block_limit=500,
+ context_window_tokens=1_624,
+ max_tokens=100,
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
)
@@ -957,8 +948,8 @@ async def test_changed_messages_use_local_estimate_after_reported_usage(monkeypa
initial_messages=[{"role": "user", "content": "new tool output"}],
tools=tools,
model="local-model",
- context_window_tokens=2_000,
- context_block_limit=500,
+ context_window_tokens=1_624,
+ max_tokens=100,
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
)
@@ -998,8 +989,8 @@ def test_resumed_provider_context_avoids_full_transcript_estimate(monkeypatch):
initial_messages=[{"role": "user", "content": "pending delta"}],
tools=tools,
model="local-model",
- context_window_tokens=2_000,
- context_block_limit=500,
+ context_window_tokens=1_624,
+ max_tokens=100,
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
)
@@ -1066,8 +1057,8 @@ async def test_runner_counts_resumed_provider_state_before_dispatch(monkeypatch)
initial_messages=[current_message],
tools=tools,
model="local-model",
- context_window_tokens=2_000,
- context_block_limit=500,
+ context_window_tokens=1_624,
+ max_tokens=100,
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
provider_state=saved_state,
@@ -1082,12 +1073,12 @@ async def test_runner_counts_resumed_provider_state_before_dispatch(monkeypatch)
@pytest.mark.asyncio
@pytest.mark.parametrize(
- ("context_block_limit", "expected_budget"),
- [(500, 500), (None, 0)],
+ ("context_window_tokens", "expected_budget"),
+ [(1_624, 500), (1_000, 0)],
)
async def test_runner_refuses_locally_fitted_request_that_still_cannot_fit(
monkeypatch,
- context_block_limit,
+ context_window_tokens,
expected_budget,
):
from nanobot.agent.runner import AgentRunner
@@ -1110,8 +1101,8 @@ async def test_runner_refuses_locally_fitted_request_that_still_cannot_fit(
],
tools=tools,
model="local-model",
- context_window_tokens=1_000,
- context_block_limit=context_block_limit,
+ context_window_tokens=context_window_tokens,
+ max_tokens=100,
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
))
@@ -1142,8 +1133,8 @@ def test_snip_history_drops_orphaned_tool_results_from_trimmed_slice(monkeypatch
model="test-model",
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
- context_window_tokens=2000,
- context_block_limit=100,
+ context_window_tokens=1_224,
+ max_tokens=100,
)
monkeypatch.setattr(
@@ -1193,8 +1184,8 @@ def test_snip_history_reserves_budget_for_tool_definitions(monkeypatch):
model="test-model",
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
- context_window_tokens=2000,
- context_block_limit=500,
+ context_window_tokens=1_624,
+ max_tokens=100,
)
def _estimate(_provider, _model, estimate_messages, estimate_tools):
@@ -1581,8 +1572,8 @@ def test_snip_history_preserves_user_message_after_truncation(monkeypatch):
model="test-model",
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
- context_window_tokens=2000,
- context_block_limit=100,
+ context_window_tokens=1_224,
+ max_tokens=100,
)
# Make estimate_prompt_tokens_chain report above budget so _snip_history activates.
@@ -1639,8 +1630,8 @@ def test_snip_history_no_user_at_all_falls_back_gracefully(monkeypatch):
model="test-model",
max_iterations=1,
max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,
- context_window_tokens=2000,
- context_block_limit=100,
+ context_window_tokens=1_224,
+ max_tokens=100,
)
monkeypatch.setattr(