prosa

a work log for AI-agent sessions

prosa /ˈpɾɔ.zɐ/  noun A conversation, a chat, an informal exchange of ideas. Prosa lets you have a conversation with your own work history.

prosa turns scattered Claude Code, Codex, Cursor, Gemini, and Antigravity session histories into one local, searchable timeline. It is built around a single load-bearing question:

What did I work on in the last N days?

— and the natural follow-ups around it: where, on which projects, with which agents and models, using which tools, taking how long, costing roughly what.

What prosa is — and isn't

prosa is a local-first, offline-friendly work log. Three small Go binaries share one module and one typed contract:

• prosa — the CLI you actually use. Reads the local store by default.
• prosa-server — a thin personal API server (Postgres + S3-compatible object storage) for the cross-device view.
• prosa-panel — a server-rendered web panel that talks to the server.

It is not a chat manager, a residential TUI, an analytics warehouse, or a multi-user SaaS. The shape and the boundaries live in INTENT.md; read that file before proposing anything substantial.

Install

# macOS — Homebrew tap (auto-published on every release)
brew install c3-oss/prosa/prosa
prosa setup

# Linux + macOS — POSIX install.sh with sha256 verification
curl -fsSL https://raw.githubusercontent.com/c3-oss/prosa/master/install.sh | sh
prosa setup

# Anywhere with Node.js 22+ — npm picks the right platform binary
npm install -g @c3-oss/prosa
prosa setup

prosa setup is a short wizard: device auth, agent discovery, scheduled-sync install (LaunchAgent on macOS, systemd user timer on Linux), and the first scan. You can also drive it manually — see docs/install.md for the full matrix (including PROSA_VERSION, INSTALL_DIR, and INSTALL_BINS for install.sh).

First timeline

Today
  11:24  laptop   claude-code   prosa       "refactor sync logic"
         -> 32min

  09:02* laptop   codex         mz-iac      "setup terraform module"
         -> 18min

Yesterday
  23:55  laptop   claude-code   prosa       "intent doc"
         -> 1h12

By default prosa reads the local store, scopes to the current project when it can detect one, and shows the last 7 days. --all lifts the scope.

Commands

Command	What it does
prosa	Timeline (auto-scoped when inside a known project)
prosa --all	Timeline across every project
prosa sync	Import local agent sessions, push to the server
prosa search <query>	Full-text search across turns (FTS5 local)
prosa show <session-id>	Print the preserved raw JSONL
prosa analytics <report>	Fixed reports: sessions/tools/errors/models/projects
prosa devices list | rename | revoke	Manage known machines (cross-device)
prosa schedule install | status | uninstall	Manage the background sync job
prosa setup	Interactive first-run wizard
prosa login	Re-authenticate (sub-step of setup)

Useful flags everywhere:

• --last 12h|7d|30d — rolling window (default 7d).
• --since 2026-01-01 — anchored lower bound, UTC.
• --between 2026-01-01..2026-03-15 — closed UTC range.
• --project <name> — project filter.
• --agent claude-code|codex|cursor|gemini|antigravity|hermes — agent filter.
• --device <name> — device filter (cross-device only).
• --remote — query the server instead of the local store.
• --json — machine-readable NDJSON output.

The three time flags (--last, --since, --between) are mutually exclusive — pick one at a time.

By default data lives under ~/.local/share/prosa (XDG). Override with PROSA_HOME. Full reference: docs/usage.md.

Self-hosting (server + panel)

If you only use prosa locally, skip this. Each binary ships as its own distroless Docker image — the image name tells you what it runs:

docker compose up -d                                          # Postgres + MinIO dev stack
docker run --rm ghcr.io/c3-oss/prosa-server:latest            # server
docker run --rm ghcr.io/c3-oss/prosa-panel:latest             # panel
docker run --rm ghcr.io/c3-oss/prosa:latest --help            # CLI

Env vars, auth, dev-login bypass, OAuth, schema details: docs/self-hosting.md.

Build from source

The repo uses devbox + just. Inside devbox shell:

just build               # builds ./bin/{prosa,prosa-server,prosa-panel}
just test                # go test ./...
just quality             # docs, links, agents, and secret scanning
just ci                  # full local pipeline (tidy/gen/vet/lint/test-race/build)
just snapshot            # local GoReleaser dry-run into dist/

The Devbox shell also installs the repo's Husky hooks for commit messages, staged Markdown, agent config, and staged secret scanning.

The deeper guide — conventions, how to add a new importer, commit style — lives in docs/contributing.md.

Documentation map

INTENT.md                            philosophy, scope, trade-offs (read first)
README.md                            you are here
ROADMAP.md                           what is being worked on next
TECH_DEBT.md                         known trade-offs we've accepted
AGENTS.md                            operational guide for repo contributors
CLAUDE.md                            pointer for Claude Code agents

docs/
├── README.md                        documentation index
├── install.md                       end-user install across channels
├── usage.md                         CLI tutorial + command reference
├── self-hosting.md                  server + panel deployment
├── concepts.md                      session lifecycle, identity, MVP scope
├── contributing.md                  code conventions + adding an importer
├── agents.md                        AI agent orientation (full)
│
├── architecture/                    how the code is really structured
├── cli/                             CLI surface design (motion, rendering, screens)
├── panel/                           web panel design (screens, components, mocks)
├── sources/                         per-agent JSONL formats
└── distribution/                    homebrew, npm, install.sh, docker, release flow

For AI agents

If you are an AI agent working on this repo, orient yourself in this order:

1. INTENT.md — read it end-to-end before proposing anything.
2. AGENTS.md — operational guide (paths, commands, conventions).
3. docs/agents.md — deeper orientation, decision checklist, where each specialist agent lives.

Specialist agents and skills are in .codex/ and .claude/. They all point back to INTENT.md and docs/; if you find a path reference that doesn't exist, that is a bug — report it.

Releases

Releases are tag-driven. Pushing a v* tag runs GoReleaser and publishes:

• macOS/Linux archives for amd64 and arm64;
• checksums.txt (sha256);
• the Homebrew cask in c3-oss/homebrew-prosa;
• the npm metapackage and four platform sub-packages;
• three multi-arch Docker images at ghcr.io/c3-oss/{prosa,prosa-server,prosa-panel}:<tag> and :latest.

Maintainer runbook: docs/distribution/release.md.

License

To the extent possible under law, Caian Ertl has waived all copyright and related or neighboring rights to this work. In the spirit of freedom of information, I encourage you to fork, modify, change, share, or do whatever you like with this project! ^C ^V