AI systems hallucinate with complete confidence. Every session opens at zero. Contradictions get smoothed over in good prose.
The missing piece isn't capability — it's epistemics. The ability to know what you know, trace your reasoning, and recognize when you're wrong.
We build systems that have epistemics:
- Memory that fades without reinforcement
- Knowledge that requires evidence — verified sources, not confident assertions
- Provenance you can trace — an unbroken line from observation to belief
- Contradictions that surface — detected on write, not discovered later
| primitives | Schema library (Apache 2.0) |
| context-service | Production backend |
| mcp | MCP server for agent harnesses |
| engine | Local-first, no cloud required (Apache 2.0) |
| research | Papers and formal specifications (CC BY-SA 4.0) |
curl -fsSL https://get.engrammic.ai | shEngrammic · 2026