An immutable, versioned log for humans and AI agents. Every entry is a Markdown file with YAML frontmatter on your own disk — grep it, diff it, commit it to git. No database, no server, no cloud account.
curl -fsSL https://logbook.codicent.ai/cli/install.sh | bash logbook init logbook add "Customer prefers dark mode #memory #ux" data/entries/2026/10/25/2026-10-25T131541Z__cef85f20….md logbook list -t memory --length 1 logbook find -t memory "dark mode"
logbook login → a Codicent projectThe install script picks the right self-contained binary for your platform from GitHub Releases. No .NET SDK needed.
curl -fsSL https://logbook.codicent.ai/cli/install.sh | bash logbook init # pin a version: … | LOGBOOK_VERSION=0.8.0 bash
iwr https://logbook.codicent.ai/cli/install.ps1 | iex logbook init # installs to %USERPROFILE%\.logbook and adds it to your user PATH
# assets: logbook-linux-x64, -linux-arm64, -osx-arm64, -osx-x64, -win-x64.exe curl -Lo logbook \ https://github.com/izaxon/codicent-logbook/releases/download/logbook-v0.8.0/logbook-linux-x64 chmod +x logbook && ./logbook --version
CLI releases are tagged logbook-v* in github.com/izaxon/codicent-logbook/releases (next to the logbook-server-v* releases).
Three layers, one data model: messages with #tags, JSON data messages and immutable versions. Pick the smallest one that does the job.
Plain files on disk, driven from the shell. For scripts, one agent's memory or a personal work log. Nothing runs in the background.
One process with SQLite, an MCP server, OpenAPI, a file inbox and an optional public tunnel. For Claude Desktop, Cursor and other MCP clients.
Logbook server →Multi-user projects, apps and 100+ automation tags (#ai, #transform, #ocr…) that turn messages into pipelines. For teams and production workflows.
Same model all the way up: the CLI writes straight to a Codicent project after logbook login, and logbook data export/import moves records between instances.
Every command is a plain shell call — easy for people, trivial for agents. Run logbook, logbook data or logbook chat for full help.
Tags come from #markers in the text or -t a,b. --amend <id> writes a new version; list shows only the latest, newest first, with --from/--to and --start/--length.
JSON key-value records with --filter k=v, --format entry|json|csv|table|export and idempotent export → import between instances.
Device-flow login stores a .logbook-session in the current folder; every command then runs against that Codicent project instead of local files.
logbook add "Deployed 1.12.2 to prod #deploy #prod" echo "long notes…" | logbook add -t notes logbook add -t deploy --amend <id> "Deployed 1.12.3" logbook list -t deploy --from 2026-10-01 logbook find --body-only "rollback" logbook history <id>
logbook data set -t customer '{"name":"Acme","status":"active"}' logbook data get -t customer --filter status=active logbook data get -t customer --format table logbook data get -t customer --format export --all \ --output customers.json logbook data import customers.json --dry-run
Entries are never rewritten or deleted. Tags are folders of links, superseded versions are marked — so plain grep, ls and git log keep working.
--- id: 3f2a1b4c-… created_utc: 2026-10-25T13:15:41Z tags: [agent, memory] parent_id: 9b7e… # previous version original_id: 9b7e… # chain root --- User confirmed they prefer dark mode in all apps.
data/ ├── entries/YYYY/MM/DD/<ts>__<id>.md │ # immutable entry files ├── tags/<tag>/<ts>__<id>.md │ # links into entries/ └── superseded/<id> # marker: a newer version exists
On Windows without Developer Mode, tag links fall back to file copies — queries behave the same.
Agents forget between sessions. Any agent that can run a command — Claude Code, Codex, OpenClaw, a cron script — can write decisions down and recall them later. No API, no auth, no SDK.
## Memory
Use the `logbook` CLI as persistent memory.
- Recall first: `logbook list -t memory --length 20`
or `logbook find -t memory "<topic>"`
- Record decisions: `logbook add "<fact> #memory #<topic>"`
- Correct a fact: `logbook add --amend <id> -t memory "<new>"`
- Never edit files under data/ by hand.
tools: - name: logbook type: shell description: Persistent memory log. Immutable, tagged, versioned. commands: add: "logbook add -t {tags} {message}" list: "logbook list -t {tags}" find: "logbook find -t {tags} {pattern}" get: "logbook get {id}" history: "logbook history {id}"
Log in once and the CLI stores entries in a Codicent project instead of local files — so a team, or several agents on different machines, share one log. Entries become Codicent messages; data commands map to #data messages.
logbook login # codicent.com, project "logbook" logbook login --url https://myserver.com --project myproj logbook status # mode, URL, project, expiry logbook add "Shared note #team" logbook logout
export LOGBOOK_REMOTE_URL="https://codicent.com" export LOGBOOK_REMOTE_TOKEN="<api-token>" export LOGBOOK_PROJECT="myproject" logbook data get -t customer --format json # env vars win over .logbook-session
Remote mode searches full message content only — --body-only/--meta-only are local-mode features.
Nothing to configure for local use: the CLI walks up from the current folder to find data/entries/, like git finds .git.
| Variable | Default | Description |
|---|---|---|
LOGBOOK_ROOT | (nearest folder with data/entries/, else cwd) | Where the local log lives. |
LOGBOOK_REMOTE_URL | (none) | Codicent base URL; with a token, switches every command to remote mode. |
LOGBOOK_REMOTE_TOKEN | (none) | Bearer token for the remote project. |
LOGBOOK_PROJECT | (none) | Remote project name. |
LOGBOOK_VERSION | (newest) | Install scripts only: pin a CLI version, e.g. 0.8.0. |