Shared context without shared chaos
Govern the memory your AI team shares.
FAVA Trails is for teams whose work spans Claude Desktop, Claude Code, ChatGPT, local files, shared drives, GitHub repos, and MCP-compatible agents. It gives every remembered fact a lifecycle before it becomes shared truth.
Draft
An agent captures a finding without exposing it as team truth.
Trust gate
A reviewer checks quality, contradictions, and scope before promotion.
Shared trail
Approved context syncs through a Git-native audit trail the team controls.
Copy-paste activation
Bootstrap a shared trail in minutes.
Start with one private data repo your team controls. The MCP setup and trust gate live in the README; this gives a technical reader the first working trail.
Open the full READMEpip install fava-trails
fava-trails install-jj
fava-trails bootstrap fava-trails-data --remote https://github.com/YOUR-ORG/fava-trails-data.gitThe team problem
Teams do not usually have one AI surface. Product and engineering may use Claude Code or Cursor. Executives may stay in browser chat. Ops may prefer shared documents. Consultants may work across several clients and devices. The storage backend is not enough; the hard part is knowing what is current, trustworthy, attributable, and safe to reuse.
Avoid context silos
Agents can recall shared decisions by scope instead of depending on one machine, one chat history, or one person's local folder.
Keep trust explicit
Draft thoughts stay isolated until promoted. Bad assumptions, transient errors, and noisy notes do not become institutional memory by accident.
Track who changed what
Every thought carries provenance, scope, source type, confidence, and lineage. Supersession keeps corrections connected to the facts they replaced.
Report without copy-paste
A weekly stoplight report can read from shared trails instead of asking every operator to re-brief the reporting agent from scratch.
Where FAVA Trails fits
| Approach | Works well for | Breaks down when |
|---|---|---|
| Shared docs or drives | Human-readable libraries, templates, and working notes. | Agents overwrite, duplicate, or trust stale content without lifecycle metadata. |
| GitHub repos | Technical teams that already review changes through pull requests. | Non-technical teams need agents to write and recall context without learning Git workflows. |
| Vector memory | Fast semantic recall over a large body of content. | Old beliefs and corrections are both returned with no reliable current-truth signal. |
| FAVA Trails | Shared agent memory that needs scope, review, rollback, and provenance. | You only need a private scratchpad and do not care whether other agents reuse it. |
A practical rollout
Start with the smallest workflow where shared memory already hurts.
Create one private trail repo
Keep the data in a Git repo the team controls. Use scopes such as company/product/research,
company/ops/weekly-report, or client-name/program/risk.
Connect the first two AI surfaces
Start with the gap that creates the most copy-paste: for example Claude Desktop research and Claude Code implementation, or an executive browser workflow and an ops reporting agent.
Use draft, propose, sync as the default habit
Agents save working context as drafts, promote durable facts through the trust gate, then sync so other agents see only reviewed context by default.
Add weekly reporting after the trail is useful
Once the team trusts the trail, use lifecycle hooks and scoped recall to summarize risks, blockers, decisions, and open questions into a stoplight report.
Want this inside your team?
FAVA Trails is open source. Machine Wisdom can also help design the scope model, trust gate, reporting loop, and agent operating protocol for your team.