Solo & independent use

No company, no org server, no API key. Manthana runs entirely on your laptop and becomes a memory of your own work.

curl -LsSf https://github.com/Jarus77/manthana/releases/latest/download/install.sh | sh
manthana solo

That’s the whole setup.

What manthana solo does

It’s the counterpart to manthana setup, which needs an invite from an admin. In one command it:

1. marks this install as solo ([mode] solo = true in ~/.manthana/manthana.toml),
2. captures your existing Claude Code and Codex history,
3. compacts everything that has settled,
4. installs the capture daemon to run at login,
5. serves your personal wiki at http://127.0.0.1:8765.

✓ solo mode — Manthana runs entirely on this laptop
  captured 214 session(s) · compacted 187 · auto-capture: installed (runs at login)
  your wiki:  http://127.0.0.1:8765
  ask it:     manthana ask "what did I try for X?"
  check it:   manthana doctor

Flag	Default	Meaning
--no-service	daemon installed	Skip the login daemon; run manthana watch yourself
--no-dashboard	wiki served	Set up and exit instead of serving the wiki
--port <n>	8765	Port for the personal wiki

It refuses to run if this install is already connected to an org server, and changes nothing:

✗ this install is already connected to an org server.
  server: https://manthana.acme.com
  Solo mode is for installs with no org. Nothing was changed.

That’s deliberate. Silently flipping a team member to solo would stop their work reaching the org wiki with no visible symptom at all — no error, no warning, just colleagues who stop seeing what they’re doing. If you really do want to leave an org, disconnect first by removing the [server] block from ~/.manthana/manthana.toml.

Why [mode] solo exists

Everything on the solo path already worked. What was missing was a way to say you’re solo, so manthana doctor could tell “hasn’t finished onboarding” from “is one person and always will be”.

Without it, doctor reported a perfectly healthy personal install as two critical failures and exited non-zero — which is exactly the signal a new user reads as “this thing is broken”. With the flag set:

Manthana doctor (solo — no org server)
  ✓ configured — solo mode — everything stays on this laptop
  • no server — by design — `manthana dashboard` is your wiki
  ✓ model available (for `manthana ask`)
  ✓ auto-capture daemon installed
  • data: 214 sessions · 187 compactions (0 pending) · 0 synced · last sync never

Exit code 0. Safe to put in a health check.

Using it

manthana dashboard                                   # http://127.0.0.1:8765
manthana insights --since 7d                         # rollups; no model call
manthana ask "how did I end up solving the X bug?"   # cited answer over your work
manthana related <session-id>                        # prior work like this session
manthana mine-skills --write                         # draft skills from your patterns

manthana ask needs no API key. It runs through the claude CLI you already have installed, falling back to codex. No request is made to us at any point. Confirm it’s wired up with manthana doctor — look for ✓ model available (for manthana ask).

Everything else — capture, compaction, insights, related, mine-skills — is deterministic and calls no model at all.

Where your data is

manthana datahome
# data_home: /Users/you/.manthana
# db_path:   /Users/you/.manthana/manthana.db

One directory: the SQLite store, manthana.toml, and the update-check cache. Back it up, move it with MANTHANA_DATA_HOME, or rm -rf it to forget everything. Nothing else on your machine is touched except the login service.

There is no server, so no sync path exists at all — the release and redaction machinery still runs, but nothing has anywhere to go.

---

Going further: your own wiki server

The local dashboard is a good personal wiki. If you want the full wiki layer — living project articles, typed notes, cross-links, entity pages — run the server locally too. Still no API key required.

# 1. Install the server
curl -LsSf https://github.com/Jarus77/manthana/releases/latest/download/install.sh | sh -s server

# 2. Run it, using the Claude CLI you already pay for as the model
export MANTHANA_SERVER_LLM=claude_cli
export MANTHANA_SERVER_ENABLE_ENRICHMENT=1
export MANTHANA_SERVER_ENABLE_CONSOLIDATION=1
export MANTHANA_SERVER_ENABLE_PROJECT_OVERVIEW=1
manthana-server serve --k-anon 1
#    → prints the admin token and http://127.0.0.1:8000

# 3. In another terminal, provision yourself and connect
manthana-server enroll me solo --open --server-url http://127.0.0.1:8000
manthana setup mia_… --actor you@example.com

Open http://127.0.0.1:8000/ui and sign in with the printed admin token.

Note that step 3 connects this install to a server, so it is no longer “solo” in the manthana solo sense — and manthana solo will now refuse to run. That’s fine: it’s still entirely your laptop. It just isn’t the zero-server path any more.

MANTHANA_SERVER_LLM=claude_cli — bring your own model

The server shells out to a Claude CLI that is installed and logged in as the user running the server. You spend the Claude subscription you already have instead of buying a separate ANTHROPIC_API_KEY.

MANTHANA_SERVER_LLM=claude_cli
MANTHANA_SERVER_CLAUDE_CLI=claude     # the binary, if it isn't called `claude`

Two things to know:

• It works on a laptop and not in a container. It needs both the binary and a logged-in $HOME, so manthana-server serve run as you is fine and the published container images are not. That’s why it’s opt-in rather than an automatic fallback.
• If the binary is missing it logs loudly and falls back to the mock, which returns {}. Your server keeps running and your wiki stays honest — it just stays empty. Check the server’s log if articles never appear.

Cost is metered from what the CLI actually reports, not estimated from a price table, so the usage numbers reflect your real subscription usage.

The alternatives are unchanged: MANTHANA_SERVER_LLM=anthropic with an ANTHROPIC_API_KEY, or the default mock, which honestly returns “insufficient data” rather than inventing something. Even on mock the server stores and serves your digests faithfully — the primary sources are produced locally and cost nothing.

--k-anon 1 — read this before you copy it

The k-anonymity floor withholds cross-contributor views until at least 4 distinct people have contributed, so nobody can be re-identified inside an aggregate. With exactly one contributor there is nobody to de-identify from, and the floor only hides your own data from you.

1 is therefore correct for a genuine single-person install, and wrong for anything else. If you ever add one other person, put it back:

manthana-server serve      # defaults to 4

The server warns at startup whenever the floor is below 4.

Secrets, if you set them

If you don’t set MANTHANA_SERVER_JWT_SECRET and MANTHANA_SERVER_ADMIN_TOKEN, the server generates a pair once, persists them to ~/.manthana-server/server-secrets.toml (mode 0600), and uses the SQLite database in that same directory. That’s the easy path — just leave them unset.

If you do set both, the CLI stops using its persisted pair and its default data directory, so pin the database too, or serve and enroll will each use a ./manthana-server.db relative to wherever you ran them:

export MANTHANA_SERVER_JWT_SECRET=$(openssl rand -hex 32)
export MANTHANA_SERVER_ADMIN_TOKEN=$(openssl rand -hex 24)
export MANTHANA_SERVER_DB_URL="sqlite:///$HOME/.manthana-server/manthana-server.db"

Changing the JWT secret invalidates the token your agent already has, so set it once, before you enroll, and keep the exports in a file you source.

Keeping it local

serve binds 127.0.0.1, so nothing is exposed. To reach your own wiki from a phone or a second machine, use Tailscale rather than opening a port:

manthana-server serve --tailscale --k-anon 1

Silencing the update notice

The agent tells you on stderr when a newer version is available. It only prints to a real terminal, never in CI, never when output is piped, and it never makes a network call on the command you actually ran.

export MANTHANA_NO_UPDATE_NOTIFIER=1

Or permanently, in ~/.manthana/manthana.toml:

[update]
notifier = false

Next

→ Daily use — the commands, in more depth → How Manthana works → Self-hosting — if your “org of one” grows