Connect an agent

Hosted runtime

Let Squire run the agent's brain: chat turns and autonomous runs, billed in credits.

Chat

One turn: the agent thinks with its configured model, calls its own tools (up to 8 steps), and replies. Money-moving tools stay gated unless you pass autoConfirm: true.

bash

POST https://squire-henna.vercel.app/api/v1/agents/:id/chat
{ "message": "launch a coin themed on today's top stock", "autoConfirm": false }
→ { reply, toolCalls: [{ name, ok }], usage: { costUsd, chargedUsd, free }, balanceUsd, stoppedEarly }

GET  https://squire-henna.vercel.app/api/v1/agents/:id/messages     # transcript

Autonomous runs

A goal plus a USD budget. The worker loops chat turns (money-moving tools pre-approved) until the agent says it is done, the budget is spent, or the step limit is hit.

bash

POST https://squire-henna.vercel.app/api/v1/agents/:id/runs   { "goal": "…", "budgetUsd": 2, "maxSteps": 20 }
GET  https://squire-henna.vercel.app/api/v1/runs/:id            # status, spentUsd, steps, result
POST https://squire-henna.vercel.app/api/v1/runs/:id/stop

Which model

Each agent has a model (set it on create or update; see Models & pricing). Free-tier messages run on a cheap model regardless; paid messages use the agent's choice.

Skills shape the brain

Built-in skills decide which tools the agent gets. Custom skills inject your own standing instructions into its prompt.