Recipes, release notes, and real-world use cases for Context Harness.
Context Harness v0.8.0 makes zvec-backed semantic search part of the native release binary story, fixes hyphenated keyword queries, and tightens CI coverage for the vector index path.
Context Harness is removing product telemetry and committing to a simpler rule: your tools should not phone home unless you explicitly ask them to.
Context Harness v0.7.0 adds the zvec vector-index sidecar, formal storage boundaries, XDG-friendly config paths, a cleaner docs hierarchy, and removes telemetry entirely.
A real-corpus benchmark points to a simple architecture: SQLite remains the canonical app store, while zvec becomes a rebuildable vector sidecar.
The filesystem connector now ingests and indexes PDF and Office documents — no extra config, just add the extensions to include_globs and sync.
PDF and Office extraction in the filesystem connector, plus observable progress during sync — no new config required.
Context Harness ships local embeddings on all six release targets — no ORT install, no musl or Intel Mac left behind.
One place to index your docs, Git repos, and runbooks—query it locally and expose it to Cursor and Claude via MCP.
The big release: a community extension registry inspired by cheat/cheat — one command, dozens of connectors and tools. Plus local embeddings on every binary and the build fixes that made it ship.
New stats, search explain, native JSON export, shell completions, and Docker support — everything you need to actually use Context Harness.
Context Harness now speaks the MCP protocol natively — connect Cursor, Claude, and other MCP clients directly via URL.
Write a 60-line Lua connector that parses a WordPress XML export and turns every post into a searchable document.
Combine a WordPress connector with a custom agent to create a conversational persona grounded in your own published writing.
A Lua connector that pulls issues from any GitHub repo and makes them searchable by your AI tools. Great for bug triage and project context.
Set up a shared Context Harness instance that indexes multiple repos, Jira, runbooks, and ADRs — so every engineer's AI assistant knows your whole stack.