Skip to content
Manthana docs

Manthana for engineers

Someone sent you an invite. Here’s what this is, what it does with your data, and the ten minutes of setup.

What it is

Section titled “What it is”

Manthana watches the transcripts your AI coding tools already write, distills each finished session into a short structured digest, and — for the sessions you’re willing to share — sends those digests to your team’s server, where they become a small wiki your team can actually read.

You get something out of it locally too: a searchable memory of your own work, “you’ve solved this before” surfacing, and cost rollups.

What it does with your data — the short version

Section titled “What it does with your data — the short version”
  • Capture is local. Sessions go into SQLite on your laptop. Nothing is uploaded at capture time.
  • Compaction is local, deterministic, and free. It spends no tokens and calls no model. It’s structure extraction over a transcript you already have.
  • Nothing syncs until it is released, and release is an opt-out window: a digest sits for 10 minutes before it becomes eligible.
  • Personal-mode sessions never sync. Ever. There is no admin override.
  • Free text is redacted before it leaves — keys, tokens, private key blocks, and a best-effort PII pass, on your machine, on the way out.
  • Your founder cannot browse your raw transcripts. They exist on the server (redacted) for citation and search, behind a drill-down that writes an audit row every time it’s used.

The full contract, including the parts that constrain your founder, is in What happens to your data. It’s worth ten minutes.

Install

Section titled “Install”
curl -LsSf https://github.com/Jarus77/manthana/releases/latest/download/install.sh | sh

Installs uv if you don’t have it, then the manthana CLI. Then paste the line your admin sent you:

manthana setup mia_…                       # an invite bound to you
manthana setup mia_… --actor you@acme.com  # a shared team invite

That one command redeems the invite, connects to your org server, installs the capture daemon so it runs at login (macOS launchd, Linux systemd --user, Windows Scheduled Task), runs a first capture, and confirms:

✓ connected as you@acme.com → https://manthana.acme.com
  captured 14 session(s) · auto-capture: installed (runs at login)
  dashboard: http://127.0.0.1:8765  ·  health check: manthana doctor

Prefer to run the daemon yourself? manthana setup … --no-service, then manthana watch in a terminal you keep open.

Verify

Section titled “Verify”
manthana doctor

Configured, server reachable, token accepted, database ready, model CLI available for manthana ask, daemon installed, and your local data counts with the last sync time. Exits non-zero if something critical is wrong. This is the first thing to run whenever anything looks off.

Then what?

Section titled “Then what?”

Nothing. Work normally. Capture is automatic; the first digests appear once a session has been quiet for 30 minutes.

When you want to look at what it has:

manthana dashboard      # http://127.0.0.1:8765

Your path

Section titled “Your path”
  1. What happens to your data — the timeline, the two windows, how to hold or hide anything.
  2. Daily use — the handful of commands worth knowing, and the dashboard.
  3. Troubleshooting — when doctor isn’t happy.

Uninstalling

Section titled “Uninstalling”

Nothing is installed system-wide beyond a CLI and a login service.

manthana service uninstall     # stop capture
uv tool uninstall manthana     # remove the CLI
rm -rf ~/.manthana             # remove your local store

Digests you already released stay on the org server; they’re the org’s record of work, like merged commits. Anything you kept personal or held was never there.

Next

Section titled “Next”

→ What happens to your data