IntelGraph AI Symphony Agent-Centric Configuration

A sophisticated Aider configuration that routes coding tasks to local Qwen models through LiteLLM, optimized for M2 16GB systems. This setup enables role-based AI assistance with multiple specialized agents for different development tasks.

What This Configuration Provides

This Aider configuration connects to a local LiteLLM proxy serving Qwen models, enabling:

Prerequisites

1. **LiteLLM proxy running locally** on port 4000 serving Qwen models

2. **Aider installed** (`pip install aider-chat`)

3. **Git repository** (configuration assumes git-tracked project)

4. **(Optional)** API keys for cloud agents: `GOOGLE_API_KEY` for Gemini, `XAI_API_KEY` for Grok

Configuration Breakdown

Core Settings

```yaml

Default agent: fast local Qwen coder

model: openai/local/llama

LiteLLM proxy endpoint

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

openai-api-key: sk-anything # Placeholder for local proxy

```

Quality & Performance

```yaml

pretty: true # Formatted output

stream: true # Real-time response streaming

auto-commits: false # Manual commit control

git: true # Git integration enabled

edit-format: diff # Show changes as diffs

show-diffs: true # Display diffs before applying

max-chat-history-tokens: 8192 # Context window limit

cache-prompts: true # Cache for faster responses

restore-chat-history: false # Don't restore old sessions (saves memory)

```

File Filtering

```yaml

ignore-globs: |

*.log

*.tmp

**/node_modules/**

**/dist/**

**/build/**

**/.vite/**

**/.next/**

**/coverage/**

rag/index/*.duckdb*

**/*.pyc

**/__pycache__/**

```

Usage Instructions

1. Create the Configuration File

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

```bash

cat > .aider.conf.yml << 'EOF'

IntelGraph AI Symphony - Agent-Centric Aider Config

model: openai/local/llama

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

openai-api-key: sk-anything

pretty: true

stream: true

auto-commits: false

git: true

edit-format: diff

show-diffs: true

max-chat-history-tokens: 8192

cache-prompts: true

restore-chat-history: false

encoding: utf-8

ignore-globs: |

*.log

*.tmp

**/node_modules/**

**/dist/**

**/build/**

**/.vite/**

**/.next/**

**/coverage/**

rag/index/*.duckdb*

**/*.pyc

**/__pycache__/**

EOF

```

2. Start Your LiteLLM Proxy

Ensure LiteLLM is running with your Qwen models configured:

```bash

Example LiteLLM startup (adjust model paths for your setup)

litellm --port 4000 --config litellm_config.yaml

```

3. Launch Aider with Default Agent

```bash

Uses default "guy" agent (architect/coder)

aider

```

4. Switch Agents for Specialized Tasks

The configuration supports role-based agent switching via command-line flags:

```bash

Architect/coder (default - fast local Qwen)

aider --model openai/local/llama

Security review (CPU-optimized variant)

aider --model openai/local/llama-cpu

Quick CI/PR tasks (smaller model)

aider --model openai/local/llama-small

Research agent (requires GOOGLE_API_KEY)

aider --model openai/gemini/1.5-pro

Power burst for complex tasks (requires XAI_API_KEY)

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

```

Agent Roles

| Agent Name | Model | Use Case | Requirements |

|------------|-------|----------|--------------|

| **guy** | `local/llama` | General coding, architecture design | Local LiteLLM + Qwen |

| **aegis** | `local/llama-cpu` | Security review, audit | Local LiteLLM + Qwen (CPU) |

| **hermes** | `local/llama-small` | Quick fixes, PR reviews | Local LiteLLM + Qwen (small) |

| **elara** | `gemini/1.5-pro` | Research, documentation | `GOOGLE_API_KEY` |

| **power** | `xai/grok-code-fast-1` | Complex refactoring | `XAI_API_KEY` |

Performance Notes

Example Workflow

```bash

1. Start coding session with default agent

aider

2. Add files to context

> /add src/components/Button.tsx

3. Request changes

> Add a disabled state with opacity 0.5

4. Review diff

Aider shows changes with show-diffs: true

5. Commit manually (auto-commits: false)

> /commit "feat: add disabled state to Button"

6. Switch to security review

> /exit

aider --model openai/local/llama-cpu

> /add src/components/Button.tsx

> Review this component for XSS and injection vulnerabilities

```

Troubleshooting

**Aider can't connect to model:**

**High memory usage:**

**Cloud agents not working:**

Advanced Customization

Add New Agent Roles

Edit your LiteLLM config to expose additional models, then call via:

```bash

aider --model openai/your-custom-route

```

Adjust Context Window

For longer conversations or larger files:

```yaml

max-chat-history-tokens: 16384 # Double the default

```

**Warning**: Higher token limits increase VRAM usage.

Enable Auto-Commits

For rapid prototyping without manual commit control:

```yaml

auto-commits: true

```

Related Resources