Your team's self-improving
memory.
Most teams run AI individually, so the work resets every session. Stash turns every run across the team into a shared, evolving asset that every agent builds on.
One-command install
bash -c "$(curl -fsSL https://raw.githubusercontent.com/Fergana-Labs/stash/main/install.sh)"
Claude Code
Cursor
Codex
OpenCode
OpenclawInstall
One command.
Automatic setup. No yaml, no manual plugin wiring. The CLI detects your agents and wires them up for you.
Use our managed service and be streaming in a minute, or self-host on your own infra if you'd rather keep every session in your own Postgres.
Why teams plateau on AI
Individual AI usage doesn't compound.
Every engineer is running Claude, Cursor, or Codex on the same repo. The insights, fixes, and gotchas from each session evaporate the moment the window closes. Next week, someone re-asks what was already answered.
Stash captures every run across the team and turns it into a shared layer your agents can query. The second time a question comes up, an agent answers it from the team's own history instead of starting from scratch. Call it a hive mind for your agents.
Questions your agent can now ask, and answer
How it works
Stream. Curate. Search.
The asset builds itself.
Every session flows into a shared store.
Prompts, tool calls, and session summaries push to your workspace’s history as they happen. Nothing to remember to save.
A curation agent turns noise into a wiki.
On SessionEnd, stash:sleep reads recent history and organizes it into notebooks with [[backlinks]] and a page graph. Sleep-time compute, not session time.
Every agent queries the whole team's work.
stash search runs a cross-resource agentic loop over files, history, notebooks, tables, and chats. Your agent answers with sources, not hallucinations.
See the memory form
Your team's brain,
actually visible.
Every session, page, and table gets embedded into one space. Stash plots them so you can see how your team's knowledge clusters, and which pages have become hubs the graph leans on.
3D embedding projection. History events, notebooks, and tables projected with PCA. Clusters form around topics — not folders.
Wiki page graph. Nodes are pages, edges are [[backlinks]]. Orange nodes are the hubs your agents keep citing.
One query, every source
Your agent asks.
Stash answers with receipts.
stash search runs an agentic loop across files, history, notebooks, tables, and chats. Every answer arrives with sources attached.
“Why did we raise the gateway rate-limit from 100 to 500?”
Answer
Sam raised it on Tue to unblock the batch-import flow. The old limit was throttling legitimate imports from Shopify partners. The change is safe because requests are authenticated and per-tenant, not global. history/sam:tue-14:22, notebooks/gateway-limits
What's inside
One team's work,
every agent's context.
Shared history
Every prompt and tool call streams to a team-wide event log. Searchable, filterable, attributable.
Wiki notebooks
Rich collaborative pages with [[backlinks]], page graph, and pgvector semantic search, curated by a sleep-time agent.
Agentic search
stash search runs a cross-resource loop over every surface in the workspace. One query, every source, with receipts.
Visualizations
See your team's memory as it forms: embedding projections, page graphs, activity timelines, and knowledge-density maps you can actually look at.
Real-time rooms
Agents and humans chat side-by-side in workspace channels. Coordinate, hand off, and unblock each other, all in one place.
Shareable pages
Publish research, reports, and dashboards as HTML anyone with a link can view. No login walls between teams.
Compound your team's
AI work.
Your team is already running agents. Stash turns those runs into a shared advantage that grows every day.
MIT · Self-hostable