Manthana documentation

Manthana turns the AI coding sessions your team is already having into a shared, searchable org wiki — without anyone writing documentation, and without anyone being surveilled.

Everything starts on the laptop. Sessions are captured locally, digested locally, and only the parts an engineer releases ever travel. Read How Manthana works once and the rest of these pages will make sense.

Pick your path

Founder or admin

Founders & admins

Setting Manthana up for a team.

Provision an org, invite engineers, set the privacy posture and AI budget, read the wiki and the weekly digest.

Engineer

Engineers

You were sent an invite.

Install in one command, know exactly what leaves your laptop and when, hold or hide anything you want to.

On your own

Solo & independent use

No company, no server, no API key.

Run manthana solo and get the whole thing locally: your own wiki, grounded questions over your own work, mined skills.

Operator

Self-hosting

Running the server for someone else.

Deploy the server + wiki client behind HTTPS, manage secrets, upgrade safely.

Reference (all audiences)

Page	What’s in it
How Manthana works	The pipeline end to end, the two-layer model, where every piece of data lives
Privacy & security model	What never leaves the laptop, redaction, k-anonymity, `privacy_mode`, audit trails
CLI reference	Every `manthana` and `manthana-server` command and flag
Environment variables	Every `MANTHANA_*` variable, its default, and what it does
Troubleshooting	Symptom → cause → fix, starting with `doctor`

The short version

your AI coding session
  ↓ captured every 5s to SQLite on your laptop         (never uploaded)
  ↓ compacted 30 min after the transcript goes quiet   (deterministic, local, free)
  ↓ auto-released after a 10 min grace window          (personal/held sessions never are)
  ↓ synced to your org server, redacted                (only released, non-personal work)
  ↓ enriched + consolidated server-side into wiki articles

Two windows, two different jobs — the 30-minute settle window decides when a session is finished; the 10-minute release window is your chance to stop it from being shared. They are explained properly in What happens to your data.

Licensing

Everything an engineer runs is Apache-2.0. The org server is AGPL-3.0-or-later. See LICENSE and NOTICE.