Frontier intelligence over 200× faster and 400× cheaper. Agentic OS for organizations of every size.
$
✓ hanzo (+ hanzo-node, same build)
✓ hanzo-mcp (+ mcp, same build)
✓ dev · gen · zt
ready → ~/.local/bin
Backed by
Language, routing, decisions and creative models.
Claude Code, Codex, Cursor and the rest, every model behind one key. Or let your agent sign itself up.
Install the Hanzo CLI
Accelerate Claude Code with Hanzo
Set up Codex, Cursor and more
No human required
// 1. Sign up with a wallet: get a challenge, sign it, verify (POST /v1/iam/web3/verify), exchange for a token.const nonce = await fetch(`${ID}/v1/iam/web3/nonce?chain=evm&address=${account.address}`).then((r) => r.json())const message = [`${nonce.data.domain} wants you to sign in with your Ethereum account:`, account.address, '', ...(nonce.data.statement ? [nonce.data.statement, ''] : []), `URI: ${nonce.data.uri}`, `Version: ${nonce.data.version}`, `Nonce: ${nonce.data.nonce}`, `Issued At: ${nonce.data.issuedAt}`].join('\n')const signature = await account.signMessage({ message })// Verify the signed challenge: a first-seen wallet gets an account and an organization of its own.const verifier = randomBytes(32).toString('base64url')const challenge = createHash('sha256').update(verifier).digest('base64url')const { data: code } = await fetch(`${ID}/v1/iam/web3/verify`, { method: 'POST', headers: { 'Content-Type': 'application/json' }, body: JSON.stringify({ clientId: CLIENT, chain: 'evm', scheme: 'secp256k1-eip191', address: account.address, message, signature, type: 'code', redirectUri: 'http://localhost:3000/callback', scope: 'openid profile email', codeChallenge: challenge, codeChallengeMethod: 'S256' }),}).then((r) => r.json())const token = await fetch(`${ID}/v1/iam/oauth/token`, { method: 'POST', headers: { 'Content-Type': 'application/x-www-form-urlencoded' }, body: new URLSearchParams({ grant_type: 'authorization_code', client_id: CLIENT, code, code_verifier: verifier, redirect_uri: 'http://localhost:3000/callback' }),}).then((r) => r.json())// 2. Get an API key (POST /v1/account/keys) that outlives the token. Shown once.const hanzo = new Hanzo({ apiKey: token.access_token, baseURL: API })const { key } = await hanzo.account.keys.create({ name: 'agent', type: 'secret' })// 3. Use AI (POST /v1/chat/completions) with the Hanzo SDK — OpenAI-compatible, one line.const client = new Hanzo({ apiKey: key, baseURL: API })const chat = await client.chat.completions.create({ model: 'zen-free', messages: [{ role: 'user', content: 'Reply with the single word: ready' }],})console.log('chat', chat.choices[0]?.message?.content, 'free left today:', chat.headers?.get?.('x-ratelimit-remaining'))// Or make a Kai decision — typed, calibrated, zero output tokens.const decision = await client.decisions.create({ model: 'kai', state: { ticket: 'The export button throws a 500 error since this morning.' }, questions: { team: { type: 'choice', instructions: 'Which team should handle this ticket?', criteria: { billing: 'charges, invoices, refunds', tech: 'bugs, errors, outages', other: 'anything else' } } },})console.log('decision', decision.answers, 'input tokens:', decision.usage.input_tokens, 'latency:', decision.latency_ms + 'ms')// 4. Top up in USDC or Card (POST /v1/billing/crypto/deposit · GET /v1/billing/crypto/options · GET /v1/billing/tier).const deposit = await client.billing.crypto.deposit({ chain: 'base', token: 'usdc', amountCents: 500 })console.log('send to', deposit.depositAddress, 'then poll', `${API}/v1/billing/crypto/deposit/${deposit.id}`)console.log('HANZO_API_KEY=' + key.slice(0, 12) + '… (store the whole key; it is shown once)')Hanzo Dev and Kai, each measured side by side.
3.02×
faster on the same coding tasks
Paired over 42 tasks both agents solved · 95% interval 2.06–4.36×.
Median time a task · less time is faster
9.6×
faster than an LLM call, over the network
Each a p50 timed at the client, network included: a Claude Sonnet 5.5 call as Claude Code saw it, over 650 calls, and a Kai decision on api.hanzo.ai, over 100. Served on one GPU with no network, its input repeated, Kai answers in 0.8 ms.
Average time & Kai speedup per call · less time is faster
Enso routes each answer to the right model. Kai answers a decision directly.
Enso vs frontier LLMs · GPQA Diamond
up to 14.2× less
per 1,000 questions, at higher accuracy
Enso Flash answers 90.6% of new GPQA Diamond questions at $3.26 per 1,000 — higher accuracy than Claude Fable 5.1 (87.9% at $46.25). Enso reaches 93.7% at $10.03, ahead of the best single model (93.1%).
Kai vs frontier LLMs
up to 530× less
than frontier LLMs, per input token
Kai bills input only; an LLM also bills every token it writes.
Per million tokens, input and output · less is cheaper
GPQA Diamond challenges AI with 198 graduate-level science questions. See how Enso balances accuracy and cost across 1,000 answers.
Dashed discs are scores reported by a lab or an independent evaluation:GLM 5.3 ↗Grok 4.7 ↗GPT-6 Sol ↗Kimi K3 ↗
Commits a day
Chat and build with Hanzo, free.
Free
No card needed
Try HanzoInstant access · Free forever
Your everyday AI development toolkit, with every model, included usage, and Hanzo Bot.
$8USD/month
Billed monthly
Try HanzoNo commitment · Cancel anytime
For serious builders running agents, shipping code, and collaborating in Hanzo Team with 3x higher limits.
$15USD/month
Billed monthly
Try HanzoNo commitment · Cancel anytime
Trusted by