Documentation menu

Start here

Introduction Getting started Write an agent Project anatomy

Understand Ayjnt

Harness engineering Two runtimes Human interfaces How agents work Host bridge

Agent capabilities

Callable methods State & SQLite Sessions & memory Scheduling Workflows Durable execution Inter-agent RPC Sub-agents Tools

Interfaces & integrations

Browser client Routing & middleware Voice Browser tools MCP Email Observability

CLI

Command overview new dev run build compile migrate deploy

API reference

Agent AgentClient Workflow classes Local & cloud Migrations Examples

Documentation

Tools

Give models small capabilities with explicit schemas, descriptions, and execution boundaries.

Start from a fresh project

This guide is self-contained. Create an empty Ayjnt project, then replace the starter agent with the files shown below. Run commands from the new project directory unless a step says otherwise.

terminal

bunx ayjnt new try-tools --empty
cd try-tools
bun install

# remove agents/alive once you are ready to add the guide's files

Two tool locations

Call agentTools(this) to receive one AI SDK ToolSet containing local agent tools plus allowed host proxies. Worker-side tools win name collisions.

File	Runtime	Good for
tools.ts	workerd	fetch, bindings, durable state, deployable services
tools.host.ts	Bun	files, shell, processes, local databases, native connectors

Design for a person to supervise

Describe the effect, not just the function name.
Validate every model-produced argument.
Confine paths and use argument arrays instead of shell interpolation.
Separate preparing an action from committing it.
Put approval immediately before consequential writes or execution.

Run it and verify the result

Give the exported tools to your model, ask for the demonstrated action, and inspect which runtime executed it.

terminal

bun run build
bun run dev

# the local app is now available at http://localhost:8787