This repository is archived as of 2026-05-31. Final release:
v0.3.1-final. The reasoning primitive has moved into the agent layer. The two parts of Nous worth preserving (risk + intervention scoring) now live in a small standalone library, decisionkit.
| If you wanted... | Use instead |
|---|---|
| Deterministic risk + intervention scoring (the durable value of Nous) | decisionkit |
| LLM-based commitment extraction | Your agent runtime (Claude Code, Codex, Hermes, Nomi, ...) with a tool definition |
| Persistent commitment / decision storage | Mnemos — claim it as a memory item |
| Multi-step reasoning loops | Agent runtime + tools + MCP into Mnemos |
| Operational orchestration | Was Olymp; also archived. The agent runtime owns the loop. |
- Reasoning has moved into the agent layer. Frontier models plus tool use plus MCP subsume what Nous was built to provide as a dedicated service. Maintaining a separate Go runtime for LLM extraction is no longer the right shape.
- The only Go consumer was Olymp, which is also being archived in the same initiative.
- The deterministic value (risk + intervention) was small and reusable — extracted into decisionkit. The 90%+ test coverage came along.
See ADR 0005 in Mnemos for the full decision record.
The full Nous source code is preserved at the v0.3.1-final tag. The
risk and intervention engines were extracted verbatim into
decisionkit (ADR 0004) so
consumers can keep using the deterministic scoring without resurrecting
the service.
Other Nous components (gRPC + HTTP transports, multi-backend storage, LLM extractor with Anthropic / OpenAI / Gemini / Bedrock / Ollama providers, adapter orchestration with circuit breakers, commitment + decision repositories) were not extracted. They exist to support a service deployment that no longer has consumers.
git clone https://github.com/felixgeelhaar/nous.git
cd nous
git checkout v0.3.1-finalThe repo remains read-only and publicly readable indefinitely.
Unchanged from the v0.3.1-final tag. See LICENSE.