Skip to content

Animus Forge

Multi-agent orchestration framework for production AI workflows.

Features

  • Workflow executor — YAML-defined multi-agent pipelines with mixins (AI, MCP, queue, graph)
  • Provider abstraction — Anthropic, OpenAI, Ollama, and more
  • Self-improvement — Analyze codebase, generate improvements, test in sandbox, create PRs
  • Budget management — Persistent token/cost tracking per workflow
  • Eval framework — Benchmark and score agent outputs
  • MCP tool execution — Bridge to Model Context Protocol servers
  • CLI + API + TUI — Multiple interfaces for workflow management

Install

pip install animus-forge

Quick Start

# Run a workflow
animus-forge run workflows/examples/build_task.yaml

# Self-improve a codebase
animus-forge self-improve run --provider ollama --path /my/project

# Analyze without making changes
animus-forge self-improve analyze --focus security

Self-Improvement Pipeline

The self-improve orchestrator runs a 10-stage workflow:

  1. Analyze — Static analysis identifies improvement opportunities
  2. Plan — AI generates an improvement plan from suggestions
  3. Safety check — Validates against protected files and change limits
  4. Snapshot — Creates rollback point before any changes
  5. Implement — AI generates code changes
  6. Sandbox test — Applies changes to temp copy, runs tests and lint
  7. Apply — Writes changes to the actual codebase
  8. Create PR — Creates a branch and pull request
  9. Human approval — Waits for merge approval (skippable with --auto-approve)
  10. Rollback — Automatic rollback if tests fail at any stage

Part of the Animus Monorepo

License

MIT