stn.
Get Started Free

How stndp works

stndp stores a handful of memory types. Knowing which is which is the whole mental model — once it clicks, the commands follow naturally. Each one is written by you, by your git activity, or by your AI agents, and all of them end up in one searchable context graph.

The memory types

  • Entry — a dated, team-visible update. A blocker is just an entry flagged blocked that @mentions whoever can clear it. Entries are what your team reads in the feed. → Standups & blockers
  • Checkpoint — your private working state (focus, why, next step, and a git snapshot). Never posted to a team, visible only to you. Captured at a context switch, reloaded with resume — by you or your agent. A quick dump is the same idea without the structure. → Checkpoints & notes
  • Decision — a durable, team-visible choice or learning, with its rationale. Searchable forever; can supersede an earlier decision to form a chain. → Decisions & search
  • Episode — the arc of one non-trivial unit of work: what you planned, what actually happened, the result, and the surprise (the plan→done delta, which is the real learning). Private while open, team-visible when you close it. → Episodes & bugs
  • Bug — a first-class symptom → cause → fix → where record, so a future teammate or agent searching "did anyone hit this?" gets an exact answer. → Episodes & bugs

Teams

Entries, decisions, episodes, and schedules belong to a team. You can be in many; each has a slug (e.g. hydra-engineering) and a numeric id. Most commands take a TEAM argument (slug or name); a few take the id. Checkpoints and notes are personal and can optionally be tagged to a team for context. See Teams & schedules.

The context graph

Around these records sits the context graph: the repos your work touches, the tracker projects (Jira/Linear/…) those repos track, the files inside them, the work items and PRs they link to, and the people involved. Records connect to graph nodes through edges (such as supersedes or touches) and external links (a Jira key or PR url). That's what lets search and file-context answer "what do we already know about this?" — read more in The context graph.

Memory that extends your agents

Your coding agent already has memory — but it's per-session and per-machine. stndp is the layer underneath: the same records are readable and writable by your teammates and by every agent you run, over MCP. A decision your agent logs this morning is there when a teammate's agent searches for it this afternoon. stndp doesn't replace your agent's memory; it gives it a shared, durable backbone.

Your context is yours

You own everything you write. stn export writes any team (or your checkpoints) to your own machine on demand, and you can delete your account and data at any time. See Configuration for exactly what the CLI stores where.