Open, local-first multi-agent platform for secure AI-assisted software development.
Ananta is built around one core idea: AI agents should not get uncontrolled full access to a project. They should work through a visible hub, clear worker roles, policy-based context control, auditable tasks and reproducible artifacts.
| Project | Purpose |
|---|---|
| Ananta | Hub, worker orchestration, operator TUI, task execution, policies and AI-assisted development workflows. |
| CodeCompass | Knowledge graph, retrieval and context-control layer for Ananta. It decides which project context is relevant and which worker is allowed to see it. |
| Website | Public landing page for Ananta and CodeCompass. |
CodeCompass is a decisive subproject of Ananta. It is the layer between project knowledge and AI execution.
It is intended to:
- index source code, documentation, TODOs, artifacts and test results
- connect files, symbols, tasks and outcomes as a project graph
- build small, explainable context bundles for workers
- preserve source references and provenance
- support local/cloud worker separation
- enforce least-privilege context sharing
The goal is not to dump the whole repository into a model. The goal is controlled, task-specific and policy-aware context.
Ananta is designed with a strong bias toward:
- local-first execution
- explicit worker roles
- default-deny thinking
- least-privilege context access
- auditable task execution
- clear separation between local models and external/cloud models
- Website: www.ananta.de
- CodeCompass page: www.ananta.de/codecompass.html
- Main repository: github.com/ananta888/ananta