Openshard

v0.4.16 makes your agents’ evidence more accurate when agents work across multiple repositories or long sessions. Token usage is no longer double-counted. Verification is more careful about what actually ran. Missing evidence stays unknown. Try it out: pip install -U openshard

Policy and permissions now actually control Openshard and its native agent runs. Organisation rules are loaded at the start. They can restrict things like models, budgets, write paths and commands. The decisions are then recorded in the Receipt and also show up on the Platform.

Openshard v0.4.8 is released. 1. Our OSN agent can now run real coding tasks. 2. File and command policy is live (allow / ask / deny). 3. Independent verification is in. 4. Evidence states are clearer (Observed / Not checked / Unavailable). Keep your coding agent. Add Openshard

Introducing the first version of the Openshard Platform. Receipts for AI coding agents, now synced from local runs into hosted history and evidence. Claude Code, Codex, Cursor, OpenCode → Receipt → sync → Platform. DM us if you're using agents on real repos. 60 sec demo ↓

Openshard v0.4.6 is live. The work you carry out on your coding agent locally goes into a local receipt and then to our hosted cloud platform in seconds. 8,950 tests passing. Production sync tested and working. Claude Code · Codex · Cursor · OpenCode github.com/openshard/opensha…

Local Receipts are becoming hosted Receipts. Today we shipped the foundations for secure, reliable Receipt sync from Openshard Core into the hosted Openshard Platform. Next stop will be final production verification, and then v0.4.6.

Openshard v0.4.5 is live. OpenCode Desktop now joins Claude Code, Codex and Cursor on the same local Receipt layer. Use your coding agent normally and let Openshard capture what happened: pip install -U openshard github.com/openshard

Same task again, this time with Codex. Codex fixed the regression, Openshard captured the run, and the result ended with the same receipt layer around the work. Claude Code. Cursor. Codex.

Same task. Same repo. Different agent. This time Cursor fixed the regression and OpenShard captured the run. One receipt layer across the tools developers already use.

Launching Receipts for AI coding agents. Claude Code fixed a real authentication regression. OpenShard captured the run and produced a receipt showing what changed and what happened. Use the coding agent you want. Keep one evidence layer around the work.