IntelGraph AI Symphony Agent Config

Agent-centric Aider configuration for orchestrating multiple AI roles through local LLM routing.

Overview

This skill configures Aider to work with a local LiteLLM proxy server, enabling fast model switching between different agent roles (architect/coder, security reviewer, CI/PR assistant, research). Optimized for M2 MacBook 16GB systems with intelligent caching and performance tuning.

Instructions

When a user requests this configuration:

1. **Create the `.aider.conf.yml` file** in the project root with the following content:

```yaml

IntelGraph AI Symphony - Agent-Centric Aider Config

Default to your Qwen route (fast, low-VRAM coder)

model: openai/local/llama

Aider reads these dashed keys (not an "openai:" block)

openai-api-base: http://127.0.0.1:4000/v1

openai-api-key: sk-anything

Quality & performance settings

pretty: true

stream: true

auto-commits: false

git: true

edit-format: diff

show-diffs: true

max-chat-history-tokens: 8192

Performance tuning for M2 16GB

cache-prompts: true

restore-chat-history: false

File handling

encoding: utf-8

ignore-globs: |

*.log

*.tmp

**/node_modules/**

**/dist/**

**/build/**

**/.vite/**

**/.next/**

**/coverage/**

rag/index/*.duckdb*

**/*.pyc

**/__pycache__/**

```

2. **Explain the agent role system** to the user:

The configuration defaults to `openai/local/llama` (the "guy" architect/coder role). Users can switch roles by launching Aider with different model flags:

**Guy (Architect/Coder)**: `aider --model openai/local/llama` - Default, fast Qwen-based coding

**Aegis (Security)**: `aider --model openai/local/llama-cpu` - Security review and vulnerability analysis

**Hermes (CI/PR)**: `aider --model openai/local/llama-small` - Quick CI/PR tasks and code review

**Elara (Research)**: `aider --model openai/gemini/1.5-pro` - Research tasks (requires GOOGLE_API_KEY)

**Power Burst**: `aider --model openai/xai/grok-code-fast-1` - High-performance tasks (requires XAI_API_KEY)

3. **Verify prerequisites**:

Aider must be installed (`pip install aider-chat`)

LiteLLM proxy server running at `http://127.0.0.1:4000` with the model routes configured

Git repository initialized (config enables git integration)

For cloud models (Gemini, xAI), appropriate API keys must be set

4. **Key configuration features**:

**Local LLM routing**: All requests go through LiteLLM proxy at port 4000

**Diff-based editing**: Uses `diff` format for precise code changes

**Git integration**: Tracks changes but disables auto-commits for manual control

**Token optimization**: 8192 token chat history with prompt caching enabled

**Intelligent ignores**: Excludes build artifacts, dependencies, logs, and cache files

5. **Performance optimizations for M2 systems**:

Prompt caching enabled to reduce memory pressure

Chat history restoration disabled to save VRAM

Token limit tuned for 16GB unified memory

Streaming enabled for responsive feedback

Usage Examples

**Start default coding session (Guy role)**:

```bash

aider

```

**Security review session (Aegis role)**:

```bash

aider --model openai/local/llama-cpu src/auth/*.ts

```

**Quick PR review (Hermes role)**:

```bash

aider --model openai/local/llama-small --read-only

```

**Research mode with Gemini (Elara role)**:

```bash

export GOOGLE_API_KEY=your_key

aider --model openai/gemini/1.5-pro

```

Important Notes

This configuration assumes a LiteLLM proxy with specific model routes (`local/llama`, `local/llama-cpu`, etc.)

The `openai-api-key` is a placeholder (`sk-anything`) - local proxy typically doesn't require authentication

Auto-commits are disabled - user must manually commit changes

Cloud models (Gemini, xAI) require valid API keys set as environment variables

File ignore patterns are optimized for Node.js/Python projects - adjust for other tech stacks

Constraints

Requires LiteLLM proxy server running locally on port 4000

Model routes must be pre-configured in LiteLLM config

Optimized for systems with 16GB RAM - adjust token limits for different specs

Git integration requires an initialized git repository

IntelGraph AI Symphony Agent Config

IntelGraph AI Symphony Agent Config

Overview

Instructions

IntelGraph AI Symphony - Agent-Centric Aider Config

Default to your Qwen route (fast, low-VRAM coder)

Aider reads these dashed keys (not an "openai:" block)

Quality & performance settings

Performance tuning for M2 16GB

File handling

Usage Examples

Important Notes

Constraints

Reviews (0)