Team knowledge

One memory across every agent.

Your team runs Claude Code, Codex, and Pi. Sync those sessions into shared traces, extract what matters against a learning goal, and let any agent pull the same reviewed context.

Set up session sync Knowledge CLI
Team learning stream 3 sources

Claude Code

~/.claude/projects

Codex

~/.codex/sessions

Pi

~/.pi/agent/sessions

provon knowledge find

$ provon knowledge find "package ownership"

2 items · 96% confidence

Package boundaries stay behind

the public adapter.

evidence: 3 sessions · codex, pi

Capture once, reuse everywhere

Sessions in. Shared context out.

  1. Sync

    01

    Each teammate uploads local sessions as shared traces.

    $ provon sync ~/.claude/projects --once

  2. Set a goal

    02

    Say what is worth learning before extraction runs.

    $ provon knowledge goals create "Architecture rules"

  3. Extract

    03

    Matching decisions become items with evidence links.

    $ provon knowledge settings set-extraction true

  4. Consume

    04

    Any agent pulls the same reviewed context on demand.

    $ provon knowledge find "package ownership"

A durable memory layer

Useful to people. Predictable for agents.

Evidence linked

Each item keeps the session spans that support it.

Incremental

Content hashes skip files that have not changed.

Goal scoped

Only active goals run, so items stay judgeable.

Deterministic

Term matching returns paths, not surprise context.

Let the next agent start informed.

Sync one session and build shared memory.

Set up sync