Skip to content

Open source AI coding agent. Designed for large projects and real world tasks.

Update model documentation and add built-in model reference

Dane Schneidere0f8a3903c8b
diff --git a/docs/docs/cli-reference.md b/docs/docs/cli-reference.md
index 75989a7b..baf859d7 100644
--- a/docs/docs/cli-reference.md
+++ b/docs/docs/cli-reference.md
@@ -67,7 +67,12 @@ The REPL has a few convenient flags you can use to start it with different modes
--strong Strong pack (more capable models, higher cost and slower)
--cheap Cheap pack (less capable models, lower cost and faster)
--oss Open source pack (open source models)
- --gemini Gemini pack (Gemini 2.5 Pro Preview for planning and coding, default models for other roles)
+
+ --gemini-planner Gemini pack (Gemini 2.5 Pro for planning, default models for other roles)
+ --o3-planner OpenAI o3-medium for planning, default models for other roles
+ --r1-planner DeepSeek R1 for planning, default models for other roles
+ --perplexity-planner Perplexity for planning, default models for other roles
+ --opus-planner Anthropic Opus 4 for planning, default models for other roles
```
All commands listed below can be run in the REPL by prefixing them with a backslash (`\`), e.g. `\new`.
@@ -107,7 +112,15 @@ plandex new -n new-plan # with name
`--oss`: Start the plan with the open source model pack.
-`--gemini`: Start the plan with the Gemini model pack.
+`--gemini-planner`: Start the plan with the Gemini planner model pack.
+
+`--o3-planner`: Start the plan with the OpenAI o3-medium planner model pack.
+
+`--r1-planner`: Start the plan with the DeepSeek R1 planner model pack.
+
+`--perplexity-planner`: Start the plan with the Perplexity planner model pack.
+
+`--opus-planner`: Start the plan with the Anthropic Opus 4 planner model pack.
### plans
diff --git a/docs/docs/core-concepts/configuration.md b/docs/docs/core-concepts/configuration.md
index 99cd6da3..f4366d11 100644
--- a/docs/docs/core-concepts/configuration.md
+++ b/docs/docs/core-concepts/configuration.md
@@ -5,7 +5,7 @@ sidebar_label: Configuration
# Configuration
-Plandex v2 provides a flexible configuration system that lets you customize its behavior based on the task you're working on and your preferences.
+Plandex provides a flexible configuration system that lets you customize its behavior based on the task you're working on and your preferences.
## Viewing Config
@@ -64,10 +64,18 @@ Autonomy settings control the overall level of automation Plandex will use. See
| `auto-commit` | Commit changes to git when applied | `true` |
| `auto-revert-on-rewind` | Revert project files when rewinding | `true` |
+### Editor
+
+| Setting | Description | Default |
+| ----------------------- | ---------------------------------------- | ------- |
+| `editor` | Editor to use for editing files | |
+
+
+
## Command Line Overrides
-Settings can be overridden with command line flags:
+Many settings can be overridden with command line flags:
```bash
# this will apply changes, automatically execute commands, and automatically debug regardless of your autonomy level and config settings
diff --git a/docs/docs/core-concepts/execution-and-debugging.md b/docs/docs/core-concepts/execution-and-debugging.md
index 5f4773a8..4b9b043e 100644
--- a/docs/docs/core-concepts/execution-and-debugging.md
+++ b/docs/docs/core-concepts/execution-and-debugging.md
@@ -5,7 +5,7 @@ sidebar_label: Execution and Debugging
# Execution and Debugging
-Plandex v2 includes command execution and automated debugging capabilities that aim to balance power, control, and safety.
+Plandex includes command execution and automated debugging capabilities that aim to balance power, control, and safety.
## Command Execution
diff --git a/docs/docs/core-concepts/reviewing-changes.md b/docs/docs/core-concepts/reviewing-changes.md
index bc2d0b5c..ed753fe6 100644
--- a/docs/docs/core-concepts/reviewing-changes.md
+++ b/docs/docs/core-concepts/reviewing-changes.md
@@ -63,7 +63,7 @@ plandex apply
### Apply Flags & Config
-Plandex v2 introduces several [new config settings and flags](./configuration.md) for the `apply` command that give you control over what happens after changes are applied.
+Plandex v2 introduced several [new config settings and flags](./configuration.md) for the `apply` command that give you control over what happens after changes are applied.
### Command Execution & Debugging
diff --git a/docs/docs/models/built-in/_category_.json b/docs/docs/models/built-in/_category_.json
new file mode 100644
index 00000000..1318dcd6
--- /dev/null
+++ b/docs/docs/models/built-in/_category_.json
@@ -0,0 +1,6 @@
+{
+ "label": "Built-In",
+ "position": 8,
+ "collapsible": true,
+ "collapsed": false
+}
\ No newline at end of file
diff --git a/docs/docs/models/built-in/built-in-models.md b/docs/docs/models/built-in/built-in-models.md
new file mode 100644
index 00000000..069c74f5
--- /dev/null
+++ b/docs/docs/models/built-in/built-in-models.md
@@ -0,0 +1,472 @@
+---
+sidebar_position: 1
+sidebar_label: Models
+---
+
+# Built-In Models
+
+Plandex includes a curated selection of built-in models.
+
+## OpenAI
+
+### `openai/o3-high`
+
+- OpenAI o3 (high reasoning)
+- Max Tokens: 200k
+- Max Output: 100k
+- Reserved Output: 40k
+- Effective Input: 160k
+- Features: XML output, no system prompt, fixed parameters, reasoning effort
+- Providers: OpenAI, Azure OpenAI, OpenRouter
+
+### `openai/o3-medium`
+
+- OpenAI o3 (medium reasoning)
+- Max Tokens: 200k
+- Max Output: 100k
+- Reserved Output: 40k
+- Effective Input: 160k
+- Features: XML output, no system prompt, fixed parameters, reasoning effort
+- Providers: OpenAI, Azure OpenAI, OpenRouter
+
+### `openai/o3-low`
+
+- OpenAI o3 (low reasoning)
+- Max Tokens: 200k
+- Max Output: 100k
+- Reserved Output: 40k
+- Effective Input: 160k
+- Features: XML output, no system prompt, fixed parameters, reasoning effort
+- Providers: OpenAI, Azure OpenAI, OpenRouter
+
+### `openai/o4-mini-high`
+
+- OpenAI o4-mini (high reasoning)
+- Max Tokens: 200k
+- Max Output: 100k
+- Reserved Output: 40k
+- Effective Input: 160k
+- Features: JSON output, no system prompt, fixed parameters, reasoning effort
+- Providers: OpenAI, Azure OpenAI, OpenRouter
+
+### `openai/o4-mini-medium`
+
+- OpenAI o4-mini (medium reasoning)
+- Max Tokens: 200k
+- Max Output: 100k
+- Reserved Output: 30k
+- Effective Input: 170k
+- Features: JSON output, no system prompt, fixed parameters, reasoning effort
+- Providers: OpenAI, Azure OpenAI, OpenRouter
+
+### `openai/o4-mini-low`
+
+- OpenAI o4-mini (low reasoning)
+- Max Tokens: 200k
+- Max Output: 100k
+- Reserved Output: 20k
+- Effective Input: 180k
+- Features: JSON output, no system prompt, fixed parameters, reasoning effort
+- Providers: OpenAI, Azure OpenAI, OpenRouter
+
+### `openai/gpt-4.1`
+
+- OpenAI GPT-4.1
+- Max Tokens: 1,047,576
+- Max Output: 32,768
+- Reserved Output: 32,768
+- Effective Input: 1,014,808
+- Features: JSON output, full compatibility
+- Providers: OpenAI, Azure OpenAI, OpenRouter
+
+### `openai/gpt-4.1-mini`
+
+- OpenAI GPT-4.1 Mini
+- Max Tokens: 1,047,576
+- Max Output: 32,768
+- Reserved Output: 32,768
+- Effective Input: 1,014,808
+- Features: JSON output, full compatibility
+- Providers: OpenAI, Azure OpenAI, OpenRouter
+
+### `openai/gpt-4.1-nano`
+
+- OpenAI GPT-4.1 Nano
+- Max Tokens: 1,047,576
+- Max Output: 32,768
+- Reserved Output: 32,768
+- Effective Input: 1,014,808
+- Features: JSON output, full compatibility
+- Providers: OpenAI, Azure OpenAI, OpenRouter
+
+## Anthropic
+
+### `anthropic/claude-opus-4`
+
+- Anthropic Claude Opus 4
+- Max Tokens: 200k
+- Max Output: 128k
+- Reserved Output: 20k
+- Effective Input: 180k
+- Features: XML output, cache control, single message mode
+- Providers: Anthropic, AWS Bedrock, Google Vertex, OpenRouter
+
+### `anthropic/claude-sonnet-4`
+
+- Anthropic Claude Sonnet 4
+- Max Tokens: 200k
+- Max Output: 128k
+- Reserved Output: 40k
+- Effective Input: 160k
+- Features: XML output, cache control, single message mode
+- Providers: Anthropic, AWS Bedrock, Google Vertex, OpenRouter
+
+### `anthropic/claude-sonnet-4-thinking`
+
+- Claude Sonnet 4 (visible reasoning)
+- Max Tokens: 200k
+- Max Output: 128k
+- Reserved Output: 40k
+- Effective Input: 160k
+- Features: XML output, cache control, single message mode, reasoning budget 32k
+- Providers: Anthropic, AWS Bedrock, Google Vertex, OpenRouter
+
+### `anthropic/claude-sonnet-4-thinking-hidden`
+
+- Claude Sonnet 4 (hidden reasoning)
+- Max Tokens: 200k
+- Max Output: 128k
+- Reserved Output: 40k
+- Effective Input: 160k
+- Features: XML output, cache control, single message mode, reasoning budget 32k
+- Providers: Anthropic, AWS Bedrock, Google Vertex, OpenRouter
+
+### `anthropic/claude-3.7-sonnet`
+
+- Anthropic Claude 3.7 Sonnet
+- Max Tokens: 200k
+- Max Output: 128k
+- Reserved Output: 20k
+- Effective Input: 180k
+- Features: XML output, cache control, single message mode
+- Providers: Anthropic, AWS Bedrock, Google Vertex, OpenRouter
+
+### `anthropic/claude-3.7-sonnet-thinking`
+
+- Claude 3.7 Sonnet (visible reasoning)
+- Max Tokens: 200k
+- Max Output: 128k
+- Reserved Output: 20k
+- Effective Input: 180k
+- Features: XML output, cache control, single message mode, reasoning budget 32k
+- Providers: Anthropic, AWS Bedrock, Google Vertex, OpenRouter
+
+### `anthropic/claude-3.7-sonnet-thinking-hidden`
+
+- Claude 3.7 Sonnet (hidden reasoning)
+- Max Tokens: 200k
+- Max Output: 128k
+- Reserved Output: 20k
+- Effective Input: 180k
+- Features: XML output, cache control, single message mode, reasoning budget 32k
+- Providers: Anthropic, AWS Bedrock, Google Vertex, OpenRouter
+
+### `anthropic/claude-3.5-sonnet`
+
+- Anthropic Claude 3.5 Sonnet
+- Max Tokens: 200k
+- Max Output: 128k
+- Reserved Output: 20k
+- Effective Input: 180k
+- Features: XML output, cache control, single message mode
+- Providers: Anthropic, Google Vertex, AWS Bedrock, OpenRouter
+
+### `anthropic/claude-3.5-haiku`
+
+- Anthropic Claude 3.5 Haiku
+- Max Tokens: 200k
+- Max Output: 8,192
+- Reserved Output: 8,192
+- Effective Input: 191,808
+- Features: XML output, cache control, single message mode
+- Providers: Anthropic, Google Vertex, AWS Bedrock, OpenRouter
+
+## Google
+
+### `google/gemini-2.5-pro`
+
+- Google Gemini 2.5 Pro
+- Max Tokens: 1,048,576
+- Max Output: 65,535
+- Reserved Output: 65,535
+- Effective Input: 983,041
+- Features: XML output
+- Providers: Google AI Studio, Google Vertex, OpenRouter
+
+### `google/gemini-2.5-flash`
+
+- Google Gemini 2.5 Flash
+- Max Tokens: 1,048,576
+- Max Output: 65,535
+- Reserved Output: 65,535
+- Effective Input: 983,041
+- Features: XML output
+- Providers: Google AI Studio, Google Vertex, OpenRouter
+
+### `google/gemini-2.5-flash-thinking`
+
+- Gemini 2.5 Flash (visible reasoning)
+- Max Tokens: 1,048,576
+- Max Output: 65,535
+- Reserved Output: 65,535
+- Effective Input: 983,041
+- Features: XML output, visible reasoning
+- Providers: Google AI Studio, Google Vertex, OpenRouter
+
+### `google/gemini-2.5-flash-thinking-hidden`
+
+- Gemini 2.5 Flash (hidden reasoning)
+- Max Tokens: 1,048,576
+- Max Output: 65,535
+- Reserved Output: 65,535
+- Effective Input: 983,041
+- Features: XML output, hidden reasoning
+- Providers: Google AI Studio, Google Vertex, OpenRouter
+
+### `google/gemini-pro-1.5`
+
+- Google Gemini 1.5 Pro
+- Max Tokens: 2M
+- Max Output: 8,192
+- Reserved Output: 8,192
+- Effective Input: 1,991,808
+- Features: XML output
+- Providers: Google AI Studio, Google Vertex, OpenRouter
+
+## DeepSeek
+
+### `deepseek/v3`
+
+- DeepSeek V3
+- Max Tokens: 64k
+- Max Output: 8,192
+- Reserved Output: 8,192
+- Effective Input: 55,808
+- Features: XML output
+- Providers: DeepSeek, OpenRouter
+
+### `deepseek/r1`
+
+- DeepSeek R1 (visible reasoning)
+- Max Tokens: 164k
+- Max Output: 33k
+- Reserved Output: 20k
+- Effective Input: 144k
+- Features: XML output, visible reasoning
+- Providers: DeepSeek, OpenRouter
+
+### `deepseek/r1-hidden`
+
+- DeepSeek R1 (hidden reasoning)
+- Max Tokens: 164k
+- Max Output: 33k
+- Reserved Output: 20k
+- Effective Input: 144k
+- Features: XML output, hidden reasoning
+- Providers: DeepSeek, OpenRouter
+
+### `deepseek/r1-70b`
+
+- DeepSeek R1 70B (Ollama only)
+- Max Tokens: 131,072
+- Max Output: 131,072
+- Reserved Output: 20k
+- Effective Input: 111,072
+- Features: XML output
+- Providers: Ollama
+
+### `deepseek/r1-32b`
+
+- DeepSeek R1 32B (Ollama only)
+- Max Tokens: 131,072
+- Max Output: 131,072
+- Reserved Output: 20k
+- Effective Input: 111,072
+- Features: XML output
+- Providers: Ollama
+
+### `deepseek/r1-14b`
+
+- DeepSeek R1 14B (Ollama only)
+- Max Tokens: 131,072
+- Max Output: 131,072
+- Reserved Output: 20k
+- Effective Input: 111,072
+- Features: XML output
+- Providers: Ollama
+
+### `deepseek/r1-8b`
+
+- DeepSeek R1 8B (Ollama only)
+- Max Tokens: 131,072
+- Max Output: 131,072
+- Reserved Output: 20k
+- Effective Input: 111,072
+- Features: XML output
+- Providers: Ollama
+
+## Perplexity
+
+### `perplexity/r1-1776`
+
+- Perplexity R1-1776 (visible reasoning)
+- Max Tokens: 128k
+- Max Output: 128k
+- Reserved Output: 30k
+- Effective Input: 98k
+- Features: XML output, visible reasoning
+- Providers: Perplexity, OpenRouter
+
+### `perplexity/r1-1776-hidden`
+
+- Perplexity R1-1776 (hidden reasoning)
+- Max Tokens: 128k
+- Max Output: 128k
+- Reserved Output: 30k
+- Effective Input: 98k
+- Features: XML output, hidden reasoning
+- Providers: Perplexity, OpenRouter
+
+### `perplexity/r1-1776-70b`
+
+- Perplexity R1-1776 70B (Ollama only)
+- Max Tokens: 131,072
+- Max Output: 131,072
+- Reserved Output: 20k
+- Effective Input: 111,072
+- Features: XML output
+- Providers: Ollama
+
+### `perplexity/sonar-reasoning`
+
+- Perplexity Sonar Reasoning (visible reasoning)
+- Max Tokens: 127k
+- Max Output: 127k
+- Reserved Output: 30k
+- Effective Input: 97k
+- Features: XML output, visible reasoning
+- Providers: Perplexity, OpenRouter
+
+### `perplexity/sonar-reasoning-hidden`
+
+- Perplexity Sonar Reasoning (hidden reasoning)
+- Max Tokens: 127k
+- Max Output: 127k
+- Reserved Output: 30k
+- Effective Input: 97k
+- Features: XML output, hidden reasoning
+- Providers: Perplexity, OpenRouter
+
+## Qwen
+
+### `qwen/qwen-2.5-coder-32b-instruct`
+
+- Qwen 2.5 Coder 32B
+- Max Tokens: 128k
+- Max Output: 8,192
+- Reserved Output: 8,192
+- Effective Input: 119,808
+- Features: XML output
+- Providers: OpenRouter
+
+### `qwen/qwen3-235b-local`
+
+- Qwen 3-235B (Ollama only)
+- Max Tokens: 40,960
+- Max Output: 40,960
+- Reserved Output: 8,192
+- Effective Input: 32,768
+- Features: XML output
+- Providers: Ollama
+
+### `qwen/qwen3-235b-a22b-cloud`
+
+- Qwen 3-235B A22B
+- Max Tokens: 40,960
+- Max Output: 40,960
+- Reserved Output: 8,192
+- Effective Input: 32,768
+- Features: XML output
+- Providers: OpenRouter
+
+### `qwen/qwen3-32b-local`
+
+- Qwen 3-32B (Ollama only)
+- Max Tokens: 40,960
+- Max Output: 40,960
+- Reserved Output: 8,192
+- Effective Input: 32,768
+- Features: XML output
+- Providers: Ollama
+
+### `qwen/qwen3-32b-cloud`
+
+- Qwen 3-32B
+- Max Tokens: 40,960
+- Max Output: 40,960
+- Reserved Output: 8,192
+- Effective Input: 32,768
+- Features: XML output
+- Providers: OpenRouter
+
+### `qwen/qwen3-14b-local`
+
+- Qwen 3-14B (Ollama only)
+- Max Tokens: 40,960
+- Max Output: 40,960
+- Reserved Output: 8,192
+- Effective Input: 32,768
+- Features: XML output
+- Providers: Ollama
+
+### `qwen/qwen3-14b-cloud`
+
+- Qwen 3-14B
+- Max Tokens: 40,960
+- Max Output: 40,960
+- Reserved Output: 8,192
+- Effective Input: 32,768
+- Features: XML output
+- Providers: OpenRouter
+
+### `qwen/qwen3-8b-local`
+
+- Qwen 3-8B (Ollama only)
+- Max Tokens: 32,768
+- Max Output: 32,768
+- Reserved Output: 8,192
+- Effective Input: 24,576
+- Features: XML output
+- Providers: Ollama
+
+### `qwen/qwen3-8b-cloud`
+
+- Qwen 3-8B
+- Max Tokens: 128k
+- Max Output: 20k
+- Reserved Output: 20k
+- Effective Input: 108k
+- Features: XML output
+- Providers: OpenRouter
+
+## Mistral
+
+### `mistral/devstral-small`
+
+- Mistral Devstral Small
+- Max Tokens: 128k
+- Max Output: 128k
+- Reserved Output: 16,384
+- Effective Input: 111,616
+- Features: XML output
+- Providers: Ollama, OpenRouter
\ No newline at end of file
diff --git a/docs/docs/models/built-in/built-in-packs.md b/docs/docs/models/built-in/built-in-packs.md
new file mode 100644
index 00000000..631fd2d9
--- /dev/null
+++ b/docs/docs/models/built-in/built-in-packs.md
@@ -0,0 +1,237 @@
+---
+sidebar_position: 2
+sidebar_label: Model Packs
+---
+
+# Built-In Model Packs
+
+Plandex includes a curated selection of built-in model packs that have been tested and optimized for different use cases.
+
+*A model pack is a mapping of [model roles](../models/model-roles) to [models](./built-in-models.md).*
+
+*They can also define fallback models for large context, large output, error handling, as well as a strong variant for the `builder` role.*
+
+## Core Packs
+
+### `daily-driver`
+*A mix of models from Anthropic, OpenAI, and Google that balances speed, quality, and cost. Supports up to 2M context.*
+
+- **planner** → `anthropic/claude-sonnet-4`
+ - largeContextFallback → `google/gemini-2.5-pro`
+ - largeContextFallback → `google/gemini-pro-1.5`
+- **architect** → `anthropic/claude-sonnet-4`
+ - largeContextFallback → `google/gemini-2.5-pro`
+ - largeContextFallback → `google/gemini-pro-1.5`
+- **coder** → `anthropic/claude-sonnet-4`
+ - largeContextFallback → `openai/gpt-4.1`
+- **summarizer** → `openai/o4-mini-low`
+- **builder** → `openai/o4-mini-medium`
+ - strongModel → `openai/o4-mini-high`
+- **wholeFileBuilder** → `openai/o4-mini-medium`
+- **names** → `openai/gpt-4.1-mini`
+- **commitMessages** → `openai/gpt-4.1-mini`
+- **autoContinue** → `openai/o4-mini-low`
+
+### `reasoning`
+*Like the daily driver, but uses sonnet-4-thinking with reasoning enabled for planning and coding. Supports up to 160k input context.*
+
+- **planner** → `anthropic/claude-sonnet-4-thinking-hidden`
+- **architect** → Uses planner model
+- **coder** → `anthropic/claude-sonnet-4-thinking-hidden`
+- **summarizer** → `openai/o4-mini-low`
+- **builder** → `openai/o4-mini-medium`
+ - strongModel → `openai/o4-mini-high`
+- **wholeFileBuilder** → `openai/o4-mini-medium`
+- **names** → `openai/gpt-4.1-mini`
+- **commitMessages** → `openai/gpt-4.1-mini`
+- **autoContinue** → `openai/o4-mini-low`
+
+### `strong`
+*For difficult tasks where slower responses and builds are ok. Uses o3-high for architecture and planning, claude-sonnet-4 thinking for implementation. Supports up to 160k input context.*
+
+- **planner** → `openai/o3-high`
+- **architect** → `openai/o3-high`
+- **coder** → `anthropic/claude-sonnet-4-thinking-hidden`
+- **summarizer** → `openai/o4-mini-low`
+- **builder** → `openai/o4-mini-high`
+- **wholeFileBuilder** → `openai/o4-mini-high`
+- **names** → `openai/gpt-4.1-mini`
+- **commitMessages** → `openai/gpt-4.1-mini`
+- **autoContinue** → `openai/o4-mini-medium`
+
+### `cheap`
+*Cost-effective models that can still get the job done for easier tasks. Supports up to 160k context. Uses OpenAI's o4-mini model for planning, GPT-4.1 for coding, and GPT-4.1 Mini for lighter tasks.*
+
+- **planner** → `openai/o4-mini-medium`
+- **architect** → Uses planner model
+- **coder** → `openai/gpt-4.1`
+- **summarizer** → `openai/gpt-4.1-mini`
+- **builder** → `openai/o4-mini-low`
+- **wholeFileBuilder** → `openai/o4-mini-low`
+- **names** → `openai/gpt-4.1-mini`
+- **commitMessages** → `openai/gpt-4.1-mini`
+- **autoContinue** → `openai/o4-mini-low`
+
+### `oss`
+*An experimental mix of the best open source models for coding. Supports up to 144k context, 33k per file.*
+
+- **planner** → `deepseek/r1`
+- **architect** → Uses planner model
+- **coder** → `deepseek/v3`
+- **summarizer** → `deepseek/r1-hidden`
+- **builder** → `deepseek/r1-hidden`
+- **wholeFileBuilder** → `deepseek/r1-hidden`
+- **names** → `qwen/qwen3-8b-cloud`
+- **commitMessages** → `qwen/qwen3-8b-cloud`
+- **autoContinue** → `deepseek/r1-hidden`
+
+## Provider Packs
+
+
+### `openai`
+*OpenAI blend. Supports up to 1M context. Uses OpenAI's GPT-4.1 model for heavy lifting, GPT-4.1 Mini for lighter tasks.*
+
+- **planner** → `openai/gpt-4.1`
+- **architect** → Uses planner model
+- **coder** → Uses planner model
+- **summarizer** → `openai/o4-mini-low`
+- **builder** → `openai/o4-mini-medium`
+ - strongModel → `openai/o4-mini-high`
+- **wholeFileBuilder** → `openai/o4-mini-medium`
+- **names** → `openai/gpt-4.1-mini`
+- **commitMessages** → `openai/gpt-4.1-mini`
+- **autoContinue** → `openai/o4-mini-low`
+
+### `anthropic`
+*Anthropic blend. Supports up to 180k context. Uses Claude Sonnet 4 for heavy lifting, Claude 3 Haiku for lighter tasks.*
+
+- **planner** → `anthropic/claude-sonnet-4`
+- **architect** → Uses planner model
+- **coder** → `anthropic/claude-sonnet-4`
+- **summarizer** → `anthropic/claude-3.5-haiku`
+- **builder** → `anthropic/claude-sonnet-4`
+- **wholeFileBuilder** → `anthropic/claude-sonnet-4`
+- **names** → `anthropic/claude-3.5-haiku`
+- **commitMessages** → `anthropic/claude-3.5-haiku`
+- **autoContinue** → `anthropic/claude-sonnet-4`
+
+### `gemini-planner`
+*Uses Gemini 2.5 Pro for planning, default models for other roles. Supports up to 1M input context.*
+
+- **planner** → `google/gemini-2.5-pro`
+- **architect** → Uses planner model
+- **coder** → `anthropic/claude-sonnet-4`
+ - largeContextFallback → `openai/gpt-4.1`
+- **summarizer** → `openai/o4-mini-low`
+- **builder** → `openai/o4-mini-medium`
+ - strongModel → `openai/o4-mini-high`
+- **wholeFileBuilder** → `openai/o4-mini-medium`
+- **names** → `openai/gpt-4.1-mini`
+- **commitMessages** → `openai/gpt-4.1-mini`
+- **autoContinue** → `openai/o4-mini-low`
+
+### `opus-planner`
+*Uses Claude Opus 4 for planning, default models for other roles. Supports up to 180k input context.*
+
+- **planner** → `anthropic/opus-4`
+- **architect** → Uses planner model
+- **coder** → `anthropic/claude-sonnet-4`
+ - largeContextFallback → `openai/gpt-4.1`
+- **summarizer** → `openai/o4-mini-low`
+- **builder** → `openai/o4-mini-medium`
+ - strongModel → `openai/o4-mini-high`
+- **wholeFileBuilder** → `openai/o4-mini-medium`
+- **names** → `openai/gpt-4.1-mini`
+- **commitMessages** → `openai/gpt-4.1-mini`
+- **autoContinue** → `openai/o4-mini-low`
+
+### `o3-planner`
+*Uses OpenAI o3-medium for planning, default models for other roles. Supports up to 160k input context.*
+
+- **planner** → `openai/o3-medium`
+- **architect** → Uses planner model
+- **coder** → `anthropic/claude-sonnet-4`
+ - largeContextFallback → `openai/gpt-4.1`
+- **summarizer** → `openai/o4-mini-low`
+- **builder** → `openai/o4-mini-medium`
+ - strongModel → `openai/o4-mini-high`
+- **wholeFileBuilder** → `openai/o4-mini-medium`
+- **names** → `openai/gpt-4.1-mini`
+- **commitMessages** → `openai/gpt-4.1-mini`
+- **autoContinue** → `openai/o4-mini-low`
+
+### `r1-planner`
+*Uses DeepSeek R1 for planning, Qwen for light tasks, and default models for implementation. Supports up to 56k input context.*
+
+- **planner** → `deepseek/r1`
+- **architect** → Uses planner model
+- **coder** → `anthropic/claude-sonnet-4`
+- **summarizer** → `openai/o4-mini-low`
+- **builder** → `openai/o4-mini-medium`
+- **wholeFileBuilder** → `openai/o4-mini-low`
+- **names** → `openai/gpt-4.1-mini`
+- **commitMessages** → `openai/gpt-4.1-mini`
+- **autoContinue** → `openai/o4-mini-medium`
+
+### `perplexity-planner`
+*Uses Perplexity Sonar for planning, Qwen for light tasks, and default models for implementation. Supports up to 97k input context.*
+
+- **planner** → `perplexity/sonar-reasoning`
+- **architect** → Uses planner model
+- **coder** → `anthropic/claude-sonnet-4`
+- **summarizer** → `openai/o4-mini-low`
+- **builder** → `openai/o4-mini-medium`
+- **wholeFileBuilder** → `openai/o4-mini-low`
+- **names** → `openai/gpt-4.1-mini`
+- **commitMessages** → `openai/gpt-4.1-mini`
+- **autoContinue** → `openai/o4-mini-medium`
+
+## Local Packs
+
+### `ollama`
+*Ollama experimental local blend. Supports up to 110k context. For now, more for experimentation and benchmarking than getting work done.*
+
+- **localProvider** → `ollama`
+- **planner** → `qwen/qwen3-32b-local`
+- **architect** → Uses planner model
+- **coder** → Uses planner model
+- **summarizer** → `mistral/devstral-small`
+- **builder** → `mistral/devstral-small`
+- **wholeFileBuilder** → `mistral/devstral-small`
+- **names** → `qwen/qwen3-8b-local`
+- **commitMessages** → `qwen/qwen3-8b-local`
+- **autoContinue** → `mistral/devstral-small`
+
+### `ollama-daily`
+*Ollama adaptive/daily-driver blend. Uses 'daily-driver' for heavy lifting, local models for lighter tasks.*
+
+- **localProvider** → `ollama`
+- **planner** → `anthropic/claude-sonnet-4`
+ - largeContextFallback → `google/gemini-2.5-pro`
+ - largeContextFallback → `google/gemini-pro-1.5`
+- **architect** → `anthropic/claude-sonnet-4`
+ - largeContextFallback → `google/gemini-2.5-pro`
+ - largeContextFallback → `google/gemini-pro-1.5`
+- **coder** → `anthropic/claude-sonnet-4`
+ - largeContextFallback → `openai/gpt-4.1`
+- **summarizer** → `mistral/devstral-small`
+- **builder** → `openai/o4-mini-medium`
+ - strongModel → `openai/o4-mini-high`
+- **wholeFileBuilder** → `openai/o4-mini-medium`
+- **names** → `qwen/qwen3-8b-local`
+- **commitMessages** → `qwen/qwen3-8b-local`
+- **autoContinue** → `openai/o4-mini-low`
+
+### `ollama-oss`
+*Ollama adaptive/oss blend. Uses local models for planning and context selection, open source cloud models for implementation and file edits. Supports up to 110k context.*
+
+- **localProvider** → `ollama`
+- **planner** → `deepseek/r1`
+- **architect** → Uses planner model
+- **coder** → `deepseek/v3`
+- **summarizer** → `mistral/devstral-small`
+- **builder** → `deepseek/r1-hidden`
+- **wholeFileBuilder** → `deepseek/r1-hidden`
+- **names** → `qwen/qwen3-8b-local`
+- **commitMessages** → `qwen/qwen3-8b-local`
+- **autoContinue** → `deepseek/r1-hidden`
diff --git a/docs/docs/models/ollama.md b/docs/docs/models/ollama.md
index 4b6f51f4..12f079a1 100644
--- a/docs/docs/models/ollama.md
+++ b/docs/docs/models/ollama.md
@@ -1,11 +1,12 @@
---
-sidebar_position: 6
+sidebar_position: 7
sidebar_label: Ollama Quickstart
---
# Ollama Quickstart
-Plandex works with [Ollama](https://ollama.com/) models. To use them, you need to [self-host Plandex.](../hosting/self-hosting/local-mode-quickstart.md) Ollama isn't supported with Plandex Cloud.
+Plandex works with [Ollama](https://ollama.com/) models. To use them, you need to [self-host Plandex.](../hosting/self-hosting/local-mode-quickstart.md)
+**Ollama isn't supported with Plandex Cloud.**
## Disclaimer
@@ -13,13 +14,23 @@ While local models are supported via Ollama, small models that can be run locall
The strongest open source models _are_ capable enough for decent results, but these models are quite large for running locally without a very powerful system. This isn't meant to discourage experimentation with local models, but to set expectations for what is achievable.
-To help bridge the gap as local models continue to improve their capabilities, a built-in `ollama-adaptive` model pack is available. This model pack uses local Ollama models for less demanding roles, plus larger remote models for heavy-lifting. There's also a built-in `ollama-experimental` model pack that uses local models for all roles—this is recommended for testing and benchmarking, but not for getting real work done.
+To help bridge the gap as local models continue to improve their capabilities, two 'adaptive' model packs are available: `ollama-oss` and `ollama-daily`. These model packs use local Ollama models for less demanding roles, plus larger remote models for heavy-lifting (open source models for the `oss` variant, and the same models used in the default `daily-driver` model pack for the `daily` variant).
+
+Over time, as local models improve, these adaptive model packs will be updated to use local models for more roles.
+
+There's also a built-in experimental `ollama` model pack that uses local models for all roles—this is recommended for testing and benchmarking, but not (yet) for getting real work done.
+
+## System requirements
+
+To use Ollama models, you need enough system resources to run the models you want to use. To use the built-in experimental `ollama` model pack (with qwen3:32b as the largest model), at least 32GB of RAM is recommended as an absolute minimum—48GB or more is recommended for breathing room. For the `ollama-daily` and `ollama-oss` model packs (with devstral:24b as the largest model), at least 16GB of RAM is recommended as an absolute minimum—24GB or more is recommended for breathing room.
+
+If you use [custom models and a custom model pack](./custom-models.md), you'll have full flexibility to choose the appropriate models for your system. Just remember that running Plandex prompts successfully is a challenge for even the largest local models.
## Install and run Ollama
[Download and install ollama](https://ollama.com/download) for your platform.
-Then make sure the ollama server is running:
+Then make sure the Ollama server is running:
```bash
ollama serve
@@ -27,12 +38,21 @@ ollama serve
## Pull Ollama models
-Pull the models you want to use. For the built-in `ollama-adaptive` and `ollama-experimental` model packs, pull the following models:
+Pull the models you want to use.
+
+For the built-in `ollama` model pack, pull the following models:
```bash
-ollama pull qwen3:32b
ollama pull qwen3:8b
ollama pull qwen3:14b
+ollama pull qwen3:32b
+ollama pull devstral:24b
+```
+
+For the `ollama-daily` and `ollama-oss` model packs, pull the following models:
+
+```bash
+ollama pull qwen3:8b
ollama pull devstral:24b
```
@@ -40,18 +60,20 @@ ollama pull devstral:24b
### Built-in model packs
-To use one of the built-in Ollama model packs in Plandex, decide whether you want to use `ollama-experimental`, which uses local models for all roles, but may struggle in practice, or `ollama-adaptive`, which uses local models for less demanding roles, plus the default Plandex models for heavy-lifting.
+To use one of the built-in Ollama model packs in Plandex, decide whether you want to use `ollama`, which uses local models for all roles, but may struggle in practice, `ollama-daily`, which uses local models for less demanding roles, plus the default Plandex models from the `daily-driver` model pack for heavy-lifting, or `ollama-oss`, which uses local models for less demanding roles, plus the open source Plandex models from the `oss` model pack for heavy-lifting.
```bash
-\set-model ollama-experimental # REPL
-plandex set-model ollama-experimental # CLI
+\set-model ollama # REPL
+plandex set-model ollama # CLI
```
Or:
```bash
-\set-model ollama-adaptive # REPL
-plandex set-model ollama-adaptive # CLI
+\set-model ollama-daily # REPL
+\set-model ollama-oss
+plandex set-model ollama-daily # CLI
+plandex set-model ollama-oss
```
### Custom models and model packs
@@ -83,3 +105,6 @@ When configuring a custom model pack to use Ollama, set the top-level `localProv
}
```
+## Contributors
+
+If you experiment with Plandex and Ollama and you find model combinations that work better than the built-in model packs, please chime in on [Discord](https://discord.gg/plandex-ai) or [open a PR](https://github.com/plandex-ai/plandex/pulls). The world of local models moves fast, and we can't always keep up with the cutting edge ourselves, so it would be great to have community help on filling in the gaps. With your help, we'd love to get to a place where Plandex can work effectively with 100% local models.
\ No newline at end of file
diff --git a/docs/docs/repl.md b/docs/docs/repl.md
index b7b3c4f7..13b2df2a 100644
--- a/docs/docs/repl.md
+++ b/docs/docs/repl.md
@@ -49,5 +49,11 @@ The REPL has a few convenient flags you can use to start it with different modes
--strong Strong pack (more capable models, higher cost and slower)
--cheap Cheap pack (less capable models, lower cost and faster)
--oss Open source pack (open source models)
- --gemini-exp Gemini experimental pack (Gemini 2.5 Pro Experimental for planning and coding, default models for other roles)
+
+ --gemini-planner Gemini pack (Gemini 2.5 Pro for planning, default models for other roles)
+ --o3-planner OpenAI o3-medium for planning, default models for other roles
+ --r1-planner DeepSeek R1 for planning, default models for other roles
+ --perplexity-planner Perplexity for planning, default models for other roles
+ --opus-planner Anthropic Opus 4 for planning, default models for other roles
+
```