saxenauts
Every AI conversation starts cold.
Core trick
Reads local Claude Code, ChatGPT, GitHub, and Gmail exports into a synthesized memory served to other agents over MCP.
Description
Every AI conversation starts cold. Syke fixes this.
Syke collects your digital footprint (Claude Code sessions, ChatGPT exports, GitHub, Gmail) and uses the Agent SDK with Opus 4.6 to synthesize psyche-level context — distributed to any
AI via MCP.
Built with Opus 4.6:
- Coverage-gated perception: Agent SDK + PermissionResultDeny hooks force full cross-platform exploration before synthesis
- Multi-agent: 3 Sonnet sub-agents explore in parallel, Opus synthesizes
- Federated push: every MCP session pushes observations back — living memory
- ALMA meta-learning: strategies evolve across runs (82%→89%)
- Zero config: pip install syke && syke setup --yes auto-detects everything
257 tests, 8 MCP tools, PyPI, MIT licensed.
README excerpt
# Syke [](https://pypi.org/project/syke/) [](https://python.org) [](https://github.com/saxenauts/syke/actions/workflows/ci.yml) [](LICENSE) Syke is a local memory agent that works with your other AI agents.  It runs in the background as an ambient agent, keeps up with your work across supported local harnesses and concurrent sessions, and serves a coherent memory for other agents to rely on. It reads your local agent activity and maintains a coherent timeline of objects, intent, and progress in prose. It serves a projection as `MEMEX.md` plus a CLI interface. Your agents use `syke ask`, `syke record`, and `syke memex` in their workflow. As a self-maintaining memory agent, Syke adapts by revising its own durable memory state over time. It can also act as a sidekick for debugging, brainstorming, and research alongside your primary coding agents. Syke is deliberately experimental and incomplete. It does not try to reproduce every feature in established memory products, and pre-1.0 releases may make intentional compatibility breaks. The runtime is exercised across multiple memory environments as on