AI Agents

Chatito Agent API for builders: Bearer keys, MCP, paper-first scopes, recipes, and why agent keys never include live trade.

Chatito is infrastructure for AI agents as well as humans: market context, strategy lifecycle, practice trading, Lab invent (scoped), webhooks, MCP, and SDKs. Humans still own any real-money policy.

This page is the human orientation. Machine-readable truth lives on chatito.ai.

In the product app: Help → AI Agents (/help/#agents) is the dedicated slide for keys, checkout, and short how-tos.

Tour The Screens

1. Settings · Agent API

Settings · Agent API

Mint keys for bots

1. 1Claim code if an agent started bootstrap. Confirm scopes, claim once, secret once.
2. 2Create key with scopes you need. Default is read; live is never available.
3. 3Your keys list for revoke. Treat chk_… like passwords.

2. Standalone Checkout

app.chatito.com/checkout

Pay link for AI Lab

1. 1No dashboard chrome - logo, checkout, short product intro only.
2. 2Sign in if needed, then Connect Wallet and pay onchain from a linked wallet.
3. 3Agents should prefer catalog field billing.human_checkout over hardcoding.

3. Help · AI Agents

Help · AI Agents

In-app orientation

1. 1Open from the header Help icon, then the AI Agents section (or /help/#agents).
2. 2Start here jumps to Settings keys, checkout, and machine recipes.
3. 3Public links point at chatito.ai for API truth while you stay in the product shell.

Hard Rules for Agents

1. Practice first. Agent keys are for read, paper_trade, optional lab, and manage scopes - not silent live arming.
2. Never invent live trading as an agent capability unless product docs explicitly open it. Agent keys cannot open, activate, or revoke a Smart Policy. Heat on MetaMask is a human 6.2h Sign, not an agent verb.
3. Prefer chatito.ai markdown + OpenAPI over scraped marketing copy for product facts and prices.
4. Fail closed on auth errors; don’t retry forever on 401/403.
5. Never log chk_… secrets. Store them only in the agent host secret store.
6. kill is a soft stop. close flattens paper and hides the book (confirm=true). Live ON returns 409.
7. Share, referrals, and Challenges never auto-post.

Where to Read

REST Surface

• Base: https://app.chatito.com/api/agent/v1
• No key yet: GET /public/catalog · GET /public/status · bootstrap claim / Lab pay paths
• Auth after mint: Authorization: Bearer chk_…
• Live trading is not an agent scope in current product policy
• Paper cockpit: 4-venue list/create/pause/enable/kill/close, Trade For Me (always paper), fills/orders/positions, PATCH, calendar/statistics/charts, Research Factory, notifications/share/referrals/Challenges redeem
• After Close, books leave list_strategies. The owner still reads retained autopsy with get_research (status=closed) or GET /research/books/{id}. No restore. Spawn from a closed parent is 404.
• Research Factory books stamp factory_status / ideas_in_all. CEX/DEX All / Paper / Queued / Waitlist / Rejected / Removed chips are Strategy Factory Ideas (GET /lab/factory, MCP get_factory_ideas, read-only).
• Each research book stamps read-only provenance (origin, created_by, owner=self, system_seed, created_at). Mint-time Agent starter books (origin=agent_starter) are one paper book per venue (Late Edge, BTC DCA, WETH DCA, Wide ETH), not user-created strategies.
• Lab is venue-native JSON (Scientist / invent / Analyst / Scout / deep / loop / waitlist). Never SSE. Deep skips LP.
• If Scientist charged a credit but the client timed out, do not call lab_scientist again. Replay with GET /lab/scientist (MCP get_lab_scientist, read-only, risk=read, 0 credits). session_id on POST continues the thread and still charges. CEX and DEX store turns.
• If Analyst charged a credit but the client timed out, do not call lab_analyst again. Replay with GET /lab/analyst (MCP get_analyst_replay, read-only, top-level risk=read, credits=0, spec-only readOnlyHint). Hosted tools/list stamps those fields on the tool object. lab_analyst is risk=credit_write. lab_analyst action=reports is LP Ask Analyst reports only. It does not replay CEX/DEX chat.
• Invent (invent_strategy) spends 8 credits when it starts. CEX/DEX accept=true promotes a Lab catalog signal; it does not start a paper book and activate is ignored. CEX is spot USDT only (no perpetuals). To run paper, create_paper_strategy. If invent charged but the client timed out, do not invent again. Replay with GET /lab/invent (MCP get_lab_invent, risk=read, 0 credits). Contract: chatito.ai/agent-api.md#invent.

Exact paths and payloads: OpenAPI + agent-api.md on chatito.ai.

Scopes

How Humans Get Keys

Default mint scope is read. Add other scopes only when needed.

Pay For AI Lab

Agents and pay links should send humans to the standalone checkout (no full dashboard chrome):

Public plan targets: $19 / mo · $190 / yr. Confirm live numbers from the catalog - do not invent prices from training data.

Agent-native wallet SIWE mint (bootstrap intent: lab_month|lab_year) is documented on agent-api.md and recipes.

MCP and SDKs

Tool map and recipes: chatito.ai/recipes.md.

Dashboard Login

Wallet SIWE is great for crypto-native agents. Many humans need email magic link:

1. Pass email on bootstrap create, or
2. POST /public/bootstrap/{code}/email, or
3. With scope account:manage: POST /account/email

Then sign in at app.chatito.com with that email. Same Chatito user as wallet and keys. The Agent API does not send the magic link; the app Auth UI does.

In-App Help

First Human Checklist

1. Sign in at app.chatito.com.
2. Settings → Agent API → Create key (chk_…). Store the secret once.
3. Optional: open checkout if you need AI Lab.
4. Optional scopes: paper_trade for create/pause/kill/close/Trade For Me; lab for invent/Scientist/Analyst (credits apply).
5. Smoke: GET /api/agent/v1/me with Authorization: Bearer chk_….
6. Follow recipes for heat → paper → Trade For Me → stats.

Related

• Settings
• Billing and credits
• Safety and Access
• Edge and Research
• Reference
• Website llms.txt for crawlers