Context and memory
What the council can see, and what SAMI remembers.
The council works from three layers of context: a shared run context that every stage reads, a slice picked for the stage at hand, and recall of anything said earlier in the run.
Hybrid retrieval blends semantic, keyword and code intelligence into one project-context stream the council can draw on.
- Shared baseline context — every stage reads it
- Per-stage context slice — pulled on top into each stage
- Cross-stage recall — look back at earlier council context
What the council can see
The council reads your project through the desktop app. Every Guardian may read files; only the Writer changes them. On paid plans, each run also carries a map of your project's file paths: the paths only, never the contents.
If your workspace has a SAMI.md or AGENTS.md file at its root, the council reads it with every run as your project's conventions: build commands, style rules, boundaries. When both exist, SAMI.md is used.
Adding context yourself
Type an @-mention in the War Room composer to attach context to your task:
| Mention | Attaches |
|---|---|
@file <path> | The content of that file. Without a path, the file open in the editor. |
@folder <path> | A listing of that folder. |
@codebase | The names of your most recently changed files. |
@docs | The CLAUDE.md guidance files in your workspace. |
@git | Your current branch and last commit message. |
What SAMI remembers
- Within a run: everything said in it. Any model can look back at what another model said in an earlier stage.
- Across runs on the same project, on paid plans: the key decisions of earlier runs.
- On your machine: your War Room chat history, and a local memory file. The settings
sami.memory.maxEntries(1000 entries by default) andsami.memory.retentionDays(90 days by default) decide how much of it is kept.