IntelGraph AI Symphony - Agent-Centric Aider Config

Multi-agent Aider configuration designed for the IntelGraph AI Symphony project, featuring role-based AI agents routed through LiteLLM for specialized development tasks.

What This Does

This Aider configuration sets up a flexible multi-agent system where different AI models handle specialized roles:

**Guy** (architect/coder) - Primary development agent using local Llama

**Aegis** (security) - Security review and vulnerability analysis

**Hermes** (CI/PR) - Quick continuous integration and pull request tasks

**Elara** (research) - Deep research using Gemini 1.5 Pro

**Power Burst** - High-performance coding with xAI Grok

The configuration routes all agents through a local LiteLLM proxy at `http://127.0.0.1:4000/v1`, enabling unified access to multiple model providers while maintaining performance on resource-constrained hardware (optimized for M2 16GB).

Configuration Details

Core Settings

**Default Model**: `openai/local/llama` (primary architect/coder agent)

**API Endpoint**: Local LiteLLM proxy at `http://127.0.0.1:4000/v1`

**Edit Format**: Unified diff format with visible changes

**Chat History**: 8192 tokens (balanced for context vs. memory)

**Git Integration**: Enabled with manual commits for review

Performance Optimizations

**Prompt Caching**: Enabled to reduce redundant processing

**Streaming**: Real-time response display

**Chat History Restore**: Disabled to reduce memory overhead

**File Encoding**: UTF-8

File Exclusions

Ignores build artifacts, dependencies, logs, and cache directories:

`node_modules/`, `dist/`, `build/`, `.vite/`, `.next/`

`coverage/`, `__pycache__/`, `*.pyc`

DuckDB indices, logs, temporary files

Agent Quick-Switch Commands

Switch between specialized agents by changing the `--model` parameter:

```bash

Primary architect/coder (default)

aider --model openai/local/llama

Security review agent

aider --model openai/local/llama-cpu

Fast CI/PR agent

aider --model openai/local/llama-small

Research agent (requires GOOGLE_API_KEY)

aider --model openai/gemini/1.5-pro

High-performance agent (requires XAI_API_KEY)

aider --model openai/xai/grok-code-fast-1

```

Setup Instructions

1. **Start LiteLLM Proxy**

Ensure LiteLLM is running on port 4000 with routes configured for all agent models:

```bash

litellm --config litellm_config.yaml --port 4000

```

2. **Create Configuration File**

Save this configuration as `.aider.conf.yml` in your project root.

3. **Set Required API Keys**

For external agents (Elara, Power Burst):

```bash

export GOOGLE_API_KEY="your-gemini-key"

export XAI_API_KEY="your-xai-key"

```

4. **Launch Default Agent**

```bash

aider

```

5. **Switch Agents as Needed**

Use the quick-switch commands above to select the appropriate agent for your task.

```

Important Notes

**Manual Commits**: Auto-commits are disabled (`auto-commits: false`) to allow review before committing changes

**Memory Constraints**: Optimized for 16GB systems; adjust `max-chat-history-tokens` if needed

**LiteLLM Dependency**: Requires a running LiteLLM proxy with configured model routes

**Git Required**: This configuration expects a Git repository (`git: true`)

**Diff Visibility**: All changes are shown in diff format (`show-diffs: true`) for transparency

Constraints

Requires LiteLLM proxy running on `http://127.0.0.1:4000/v1`

External agents (Gemini, xAI) require valid API keys

Designed for Unix-like systems (tested on macOS M2)

Local models must be configured in LiteLLM routing table

IntelGraph AI Symphony Agent Config

IntelGraph AI Symphony - Agent-Centric Aider Config

What This Does

Configuration Details

Core Settings

Performance Optimizations

File Exclusions

Agent Quick-Switch Commands

Primary architect/coder (default)

Security review agent

Fast CI/PR agent

Research agent (requires GOOGLE_API_KEY)

High-performance agent (requires XAI_API_KEY)

Setup Instructions

Usage Examples

Architectural Planning (Guy)

"Design the database schema for user authentication"

Security Review (Aegis)

"Review this authentication module for security vulnerabilities"

Quick PR Review (Hermes)

"Check this PR for code style and obvious bugs"

Deep Research (Elara)

"Research best practices for OAuth 2.0 PKCE flow implementation"

Performance Critical Work (Power Burst)

"Optimize this database query for large datasets"

Important Notes

Constraints

Reviews (0)