Ares / Aries

Local multi-agent AI system for running specialised assistants on a personal machine. Ares combines local Ollama models, role-specific agents, A2A HTTP services, memory, Serena code-navigation integration, and Langfuse observability behind a terminal REPL.

What It Does

• Routes prompts through an orchestrator agent
• Provides direct agent commands for coder, thinker, runner, and Serena workflows
• Runs local models through Ollama instead of hosted model APIs
• Exposes A2A health endpoints for individual agents
• Tracks runs through a local/self-hosted Langfuse stack
• Includes a learning cycle hook for memory and behaviour refinement experiments

Architecture

Terminal REPL
    |
    v
AresSystem
    |
    +-- orchestrator agent
    +-- coder agent
    +-- thinker agent
    +-- runner agent
    +-- Serena/code-navigation agent
    |
    +-- Ollama local models
    +-- memory layer
    +-- Langfuse observability
    +-- A2A HTTP services (:8100-:8104)

Commands

/coder <task>    code-focused task execution
/thinker <task>  deeper reasoning or planning
/runner <task>   quick execution-oriented task
/serena <task>   codebase-aware navigation and edits
/learn           run the learning cycle
/quit            exit the REPL

Requirements

• Python 3.13 environment used by the paired ares-core checkout
• Ollama with the configured local models pulled
• Docker and Docker Compose for the Langfuse stack
• Local ares-core directory next to this repository

The current scripts expect this layout:

parent-directory/
  aries/
  ares-core/

Configuration

Set Langfuse keys in your shell or a local environment file before running the startup script:

export LANGFUSE_PUBLIC_KEY="pk-lf-..."
export LANGFUSE_SECRET_KEY="sk-lf-..."

Do not commit real keys. The repository intentionally does not store production or personal secrets.

Quickstart

./start.sh

The script checks or starts:

• Ollama on localhost:11434
• Langfuse on localhost:3000
• A2A agent services on ports 8100 to 8104
• the terminal REPL through main.py

Status

This is an experimental personal AI system. It is useful as a portfolio project for local AI orchestration, but it still assumes a local machine layout and model setup. Next improvements should include a one-command installer, sample config, automated tests, and Dockerised agent services.