Every command is also available as cdb. Most of the time you will not run any of these: capture and
recall are hooks, and Claude reaches the rest over MCP.
Setup
| Command | What it does |
|---|
claude-db install [--project] | Register hooks and the MCP server |
claude-db uninstall [--project] | Remove them, keeping your memory |
claude-db status | Is it wired up, and when did it last record anything |
claude-db doctor [--deep] | Resolved config; --deep proves a full round trip |
Memory
| Command | What it does |
|---|
claude-db search [--all] [--tag <name>] <query> | Search this project's memory, or every project |
claude-db remember [--key <name>] <text> | Record a rule outright, e.g. "always use pnpm here" |
claude-db forget <id> | Delete specific observations by id |
claude-db seed --from-git | Fill a cold memory from this repo's history |
claude-db stats | What this project's memory is made of |
claude-db projects | Every project with memory stored |
claude-db view [--export <file>] | See this project's memory live in the browser |
claude-db adoption | How often sessions grep vs use the memory tools |
view starts a local server only while you are looking at it, opens the browser, and stops on
Ctrl+C — no background process is left running. --export writes a standalone HTML snapshot
instead. adoption reads this project's transcripts and reports Bash commands, greps, hook fires and
memory-tool calls, so whether any of this is working is a number rather than a feeling.
Code graph
| Command | What it does |
|---|
claude-db scan [--force] | Build the code graph for this repo |
claude-db usages [--mode <m>] <symbol> | What uses a symbol: live git grep, or the code graph |
Modes are text, usages, explain and path. See Code graph for what each
one answers.
Database
| Command | What it does |
|---|
claude-db use <url> | Switch database and verify it |
claude-db sync <url> | Two-way merge with another database |
claude-db export [--all] | Dump memory as JSONL, for backup or migration |
claude-db import <file> | Load a dump back in; safe to repeat |
claude-db merge [<path>] | Move memory from an old project path onto this one |
Maintenance
| Command | What it does |
|---|
claude-db flush | Re-ingest every transcript for this project |
claude-db reembed | Re-embed everything with the current model |
claude-db prune --older-than <days> | Delete old memory (dry run without --yes) |
claude-db reset [--project] --yes | Delete memory (dry run without --yes) |
reembed is what you run after installing @xenova/transformers, so existing rows are searchable at
the new dimension rather than only new ones. flush is what you run after upgrading claude-db, since
it re-ingests transcripts through the current extractor and updates observations in place.
Configuration
~/.claude-memory/config.json holds the settings. Every key has a working default, so the file is
optional.
| Key | Default | What it does |
|---|
embeddings.provider | auto | auto uses the real model when installed, else the built-in embedder |
inject.perPrompt | true | Inject matching memory above each prompt |
inject.minOverlap | 1 | Content words a memory must share with the prompt to be injected; 0 off |
capture.summarize | off | on writes session summaries with your own Claude CLI; costs tokens |
capture.summarizeModel | opus | Model for those summaries: opus, sonnet or haiku, always low effort |
updates | notify | Update checks: auto, notify or off |
