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¶
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:
- Analyze — Static analysis identifies improvement opportunities
- Plan — AI generates an improvement plan from suggestions
- Safety check — Validates against protected files and change limits
- Snapshot — Creates rollback point before any changes
- Implement — AI generates code changes
- Sandbox test — Applies changes to temp copy, runs tests and lint
- Apply — Writes changes to the actual codebase
- Create PR — Creates a branch and pull request
- Human approval — Waits for merge approval (skippable with
--auto-approve) - Rollback — Automatic rollback if tests fail at any stage
Part of the Animus Monorepo¶
- Animus Core — exocortex engine
- Animus Quorum — coordination protocol
- Animus Bootstrap — system daemon
License¶
MIT