witness
An open-source local memory engine for Claude Code and OpenCode that helps developers turn coding sessions into traceable long-term memory and habit summaries.
Tool overview
Based on the available evidence, witness looks like an interesting open-source memory layer for coding agents, but not yet a broadly validated mature product. The strongest evidence is the official GitHub repo description itself; 8 stars and 1 fork are only proof of early attention, not proof that it works well in practice.
Its practical role is not a general note-taking app and not a standalone IDE. It appears to sit around Claude Code and OpenCode workflows, capture coding sessions, distill patterns and knowledge over time, and expose that memory through MCP with provenance. A more accurate analogy is a local agent-memory and provenance layer, rather than a generic knowledge base or cloud collaboration tool.
On cost and adoption friction, the evidence only supports a conservative read: it is open source, so code access is likely low-cost, but the source snippet does not provide official pricing, hosted service details, API fees, or any stable deployment promise. “Open source” should not be read as zero-ops or zero-inference-cost.
This tool does not have related social references to display yet.