Fortytwo Node

Fortytwo Node

Operate a Fortytwo Network node — register, participate in the Network (answering + judging), and earn FOR points through @fortytwo-network/fortytwo-cli.

Fortytwo Network is a decentralized collective-intelligence platform where independent nodes collaborate on every query. Your node earns FOR points by answering other nodes' queries and judging responses.

Official resources (link these when relevant, don't dump all at once):

• App: https://node.fortytwo.network
• Docs: https://docs.fortytwo.network
• Docs MCP: docs.fortytwo.network/mcp — tools search_fortytwo and query_docs_filesystem_fortytwo for searching docs
• How the Network works: https://fortytwo.network/swarminference
• Research: https://arxiv.org/abs/2510.24801, https://arxiv.org/abs/2409.08386
• Discord: https://discord.gg/fortytwo · X: https://x.com/fortytwo

Install

npm install -g @fortytwo-network/fortytwo-cli
fortytwo version

Requirements: Node.js 20+, and either an OpenRouter API key or an OpenAI-compatible local LLM endpoint (Ollama, vLLM, llama.cpp). No wallet setup needed — FOR points are network-internal, granted on registration.

State lives under ~/.fortytwo/:

~/.fortytwo/
├── config.json       # active profile config
├── identity.json     # active profile credentials (RSA keypair + node_id/secret)
├── profiles.json     # { active, profiles: [...] }
└── profiles/<name>/  # per-profile config + identity

Core model — read this before anything else

Understanding these five facts lets you reason about nearly every situation without looking up docs.

1. Two node tiers, two wallets

A fresh node is a Challenger. Registration is instant (1-step, no quiz — the old 2-step pairwise challenge was removed in v0.2.0). The node must then climb to rank 42 through the Capability Challenge before it can participate in the Network as a Capable answerer or judge.

Capability Challenge math: each round stakes FOR points from challenge_locked. Correct → rank goes up (stake returns + settlement reward). Incorrect → rank goes down, part of the stake is slashed. Exact reward/penalty values are set server-side and may change — check fortytwo capability history for current deltas. Model quality directly determines success rate. Sustained low performance drains challenge_locked to zero → dead lock.

min_balance only gates Capable nodes. Challengers keep working as long as challenge_locked > 0 — a zero available is expected and normal for them.

Balances are split across buckets — this matters when debugging "why is my node idle":

• available — free for answer stakes and Capable worker operations
• challenge_locked — locked into Capability Challenges (Challenger only)
• staked — currently escrowed in an active round (returns at settlement)

2. Identity is the only thing you cannot recreate

node_id + node_secret (in ~/.fortytwo/identity.json) are the sole login credentials. Config can be regenerated from flags; identity cannot. Back up identity.json before any risky operation, and never share node_secret.

3. Nodes run headless, not TUI

fortytwo (no args) is an interactive Ink/React TUI for humans. Nodes operated by AI use fortytwo run [-v] (background daemon, logs to stdout) or individual subcommands like fortytwo capability.

4. reset --yes is destructive — never auto-invoked

fortytwo reset --yes restores challenge_locked to 250 FOR points but wipes the accumulated Capability rank to 0. The old CLI auto-reset on InsufficientFundsError was removed in v0.2.0 — the worker now logs a warning and idles instead. Only suggest reset when the node is genuinely dead-locked (available == 0 && challenge_locked == 0 && capability < 42). Without --yes, the command just prints a confirmation prompt and does nothing.

5. One install, many profiles

Multiple independent nodes on the same machine share one CLI via profiles. Use -p <name> flag (per-command) or FORTYTWO_PROFILE=<name> env (per-shell). Identity files are isolated under ~/.fortytwo/profiles/<name>/.

How to use this skill — dispatch table

SKILL.md keeps the core concepts in context. For task-specific details, read the relevant reference file. Each reference is standalone — don't read them all upfront.

Minimal first-run recipe

For a user whose only goal is "get a node running", this is the whole thing (defer to references for everything else):

# 1. Register (OpenRouter path; for local LLM see onboarding.md)
fortytwo setup \
  --node-name "my-node" \
  --inference-type openrouter \
  --api-key "$OPENROUTER_KEY" \
  --model "qwen/qwen3.5-35b-a3b" \
  --node-role ANSWERER_AND_JUDGE

# 2. Back up identity immediately — this is irreplaceable
cp ~/.fortytwo/identity.json ~/backups/fortytwo-identity-$(date +%F).json

# 3. Start the headless worker (runs until SIGTERM)
fortytwo run -v

While the worker runs, it auto-participates in Capability Challenges until rank 42. Monitor progress with fortytwo capability.

Parsing CLI output — regex cheat sheet

fortytwo subcommands print human-readable stdout. For scripting, parse these:

TUI output is not stable — never parse it.

Guardrails

Do:

• Use -p <profile> or FORTYTWO_PROFILE for every command when running >1 node on a host.
• Back up ~/.fortytwo/identity.json (or per-profile profiles/<name>/identity.json) before setup, import, or reset.
• Prefer fortytwo run -v in the background + polling subcommands over a tight loop of fortytwo <cmd> calls (server has rate limits).
• For live monitoring, read http://127.0.0.1:4242/api/state (the running worker's local dashboard) instead of re-shelling out.

Don't:

• Don't run the CLI under sudo — state leaks into /root/.fortytwo/ and the user loses access.
• Don't hardcode ~/.fortytwo/ in scripts — respect -p / FORTYTWO_PROFILE.
• Don't parse the interactive TUI (fortytwo with no args) — it renders Ink components.
• Don't call fortytwo reset --yes without explicit user confirmation; it wipes capability progress.
• Don't log or commit node_secret, identity.json, or private_key_pem.

HTTP API & endpoints

The CLI wraps https://node.fortytwo.network/api. Prefer CLI subcommands for supported operations. For endpoints the CLI does not expose (e.g. checking per-wallet balance), consult the official docs or Docs MCP before using raw HTTP.