Project memory
Project memory turns what your team has done on a repository into a short, current summary that any AI coding tool loads on demand. Type /spandock in Claude Code, Codex, Gemini CLI or Qwen Code to get what was built recently and by whom, decisions and notes, open threads, known pitfalls and the areas that changed.
Project memory is on the Premium and Enterprise plans, and in the 30-day trial.
Turn it on
- On the server's dashboard, open People & computers → Computers → Client defaults, or one computer's settings, and set Project memory to on. This installs the
/spandockcommand and SpanDock's MCP entry in the AI tools found on the computer. Off removes them. - In Claude Code, inside a repository, type
/spandock. - On the server, Insights → Project memory lists your projects. Open one to read its memory, add notes, or Refresh it.
The commands
| Command | What it does |
|---|---|
/spandock or fetch | Loads the memory. Ask for more detail, or for what it contains and when it was built. |
refresh | Summarises the project now, at most once every 5 minutes per project. |
mine | Your own sessions on the project. |
team | Who else works on it, and on what, as your administrator allows. |
note | Saves a decision or note, up to 1,000 characters, with secrets removed. |
How a memory is built
A project is a git repository. The command sends the repository's remote address, normalised to host/org/repo without credentials, and the top folder's name.
The server builds each memory from the last 30 days of:
- Work insights' work items, only from people whose prompt sharing is on;
- activity counts: sessions, tokens, cost, failures, tools, models and the folders that changed (inside the repository only, so home folders never appear);
- notes people saved.
It is written by the model Work insights uses, with secrets removed from what it is sent and from its answer, or as a plain summary without a model. A memory stays around 1,000 to 2,000 tokens. Each build is a new version, and the last 20 are kept.
Schedule: active projects (with activity in the last 14 days) are refreshed when they changed: daily by default, or hourly, weekly or never.
What it never uses
- Prompt text. Raw prompts are never read.
- Secrets, which are removed before and after the model sees the inputs.
Who can read it
- A computer can read the memory of repositories it has worked on, and add notes to them. This is on by default; Safety & compliance → Access sets it for everyone or for a group.
- Team activity is set by an administrator: full (names, sessions, tools and work titles), names only, or off.
- Notes saved with
/spandockare recorded in the audit log, by field name only.
Limits
- Two repositories with the same name in different organisations share their activity records. Their memories and notes stay separate.
- Without the plan, the command answers with an explanation instead of a memory, and the schedule doesn't run.