AgentLattice/Docs
Get Started
Navigate docs

Releases

Quick Start

Get your first AI agent action governed and audited in under five minutes. No infrastructure to deploy, no configuration files to write, no meetings to schedule.

Have an existing agent? Skip to Auto-Instrumentation → — al-instrument (Python · LangChain), al-instrument-ts (TypeScript), or the /instrument skill (any framework, Claude Code) adds governance to your existing code in one command. No rewrites.

1. Sign Up

Go to agentlattice.io and sign in with GitHub. Your workspace, your first agent, and a default policy are created automatically. You land on the dashboard ready to go.

2. Copy Your API Key

Your first agent — called "Default Agent" — is already provisioned with an API key. Find it on the dashboard under your agent settings. Copy it. You will use this key to authenticate SDK calls.

Set it as an environment variable:

export AL_API_KEY="al_your_key_here"

3. Install the SDK

TypeScript (Node.js 14.17+):

npm install @agentlattice/sdk

Python:

pip install agentlattice

4. Fire Your First Action

Three lines of code. This submits a test action to AgentLattice and records it in your audit trail:

TypeScript:

import { AgentLattice } from "@agentlattice/sdk";

const al = new AgentLattice({ apiKey: process.env.AL_API_KEY });
const result = await al.execute("sdk.test");
console.log(result.status); // "executed"

Python:

from agentlattice import AgentLattice

al = AgentLattice(api_key=os.environ["AL_API_KEY"])
result = al.execute("sdk.test")
print(result.status)  # "executed"

That is it. Your action has been evaluated against your default policy, executed, and recorded in the audit trail.

5. See It in the Dashboard

Go back to your dashboard at agentlattice.io. Open the activity feed. You will see your test action:

Default Agent  |  sdk.test  |  executed  |  just now

Every action your agents take from now on will appear here — with the agent name, action type, policy decision, and timestamp.

6. Try an Approval Gate

The real power of AgentLattice is gate() — it blocks your agent's execution until a human approves the action. To see this in practice, first create a policy in the dashboard that requires approval for deploy.trigger (see the Policies guide for details).

Then use gate() in your agent code:

TypeScript:

import { AgentLattice } from "@agentlattice/sdk";

const al = new AgentLattice({ apiKey: process.env.AL_API_KEY });

// This line blocks until a human approves in the dashboard
await al.gate("deploy.trigger", {
  metadata: { environment: "production", service: "api" },
});

// Your agent code only runs after approval
await deployToProduction();

Python:

from agentlattice import AgentLattice

al = AgentLattice(api_key=os.environ["AL_API_KEY"])

# This blocks until a human approves in the dashboard
al.gate("deploy.trigger", metadata={"environment": "production", "service": "api"})

# Your agent code only runs after approval
deploy_to_production()

When this code runs, the action appears in your dashboard's approval queue. Approve it, and the agent continues. Deny it, and the SDK throws an error your agent can handle gracefully.

What Just Happened

In five minutes, you:

  1. Created an account with an auto-provisioned agent and default policy
  2. Installed the SDK
  3. Submitted an action that was evaluated, executed, and audited
  4. Saw the full audit trail in your dashboard

Your default policy is a wildcard that allows all actions with logging enabled. This means every action your agents take is recorded, giving you full visibility from day one. As you define more specific policies, you layer on authorization rules and approval gates — without changing your agent code.

Next Steps

  • Policies: Create rules that govern what your agents can do — auto-approve safe actions, require approval for risky ones, deny unauthorized access outright.
  • Use Cases: See how AgentLattice handles real enterprise workflows: DB migrations, deployments, PR merges, and incident response.
  • Dashboard: Learn what the activity feed, approval queue, and policy editor can do.
  • SDK reference: See the full SDK documentation for TypeScript and Python packages, including error handling, delegation, retry behavior, and framework integrations.