Product · Integrations

Connect your agents to 200+ tools

An AI agent is only as capable as the tools it can reach. Plug your agents into the apps, databases, and APIs your team already runs — with scoped permissions, OAuth, and a full audit trail on every call.

  • 200+ connectors
  • Scoped OAuth
  • SOC 2 audited

AI agent integrations are the bridge between an agent's reasoning and the real world. When an agent decides to "look up the account," "open a pull request," or "post to the channel," it does so by calling an integration — a tool that wraps an external API in a typed, permission-scoped function the model can invoke through tool calling.

Our catalog ships with 200+ integrations covering communication, CRM and sales, data warehouses, developer tooling, productivity, and support. Authorize a connector once and it becomes available to every agent in your workspace — including the specialist agents inside a multi-agent system. When the tool you need isn't in the catalog, register your own REST or GraphQL endpoint with the SDK and it behaves exactly like a first-party connector.

Crucially, integrations are not just "API access." Each one carries the narrowest scopes it needs, hides credentials behind an encrypted vault, and logs every call. That's what makes the difference between an agent that's a demo and one you can trust in production. To go deeper on the underlying concept, read AI agent tools.

The catalog

Integration categories for every AI agent

From sending a Slack message to querying Snowflake, here are the connector families your agents can draw on.

Communication

Let agents read and post messages, send email, and notify the right humans. Includes Slack, Gmail, Microsoft Teams, Discord, and Twilio SMS.

CRM & Sales

Read and update customer records, enrich leads, and log activity. Includes Salesforce, HubSpot, Pipedrive, and Outreach.

Data & Warehouses

Run scoped queries against your source of truth. Includes Postgres, Snowflake, BigQuery, MySQL, and Redshift with read-only roles by default.

Developer tools

Open issues, review PRs, and track work. Includes GitHub, GitLab, Jira, Linear, and Sentry for coding and ops agents.

Productivity

Read and write docs, pages, and files. Includes Notion, Google Drive, Google Sheets, Confluence, and Airtable.

Support

Triage, draft, and resolve tickets end to end. Includes Zendesk, Intercom, Freshdesk, and Help Scout.

Web & Search

Ground answers in live information. Includes web search, page fetching, and scraping connectors for retrieval-augmented agents.

Custom REST / GraphQL

Wrap any internal or third-party API as a tool. Define the schema, point it at your endpoint, and your agent can call it like any built-in.

One catalog, every agent

Connectors are shared at the workspace level. Authorize Salesforce once and your support agent, sales agent, and any orchestrator can all reach it — no re-authentication per agent, and every grant is revocable in a single place.

Under the hood

How an AI agent integration works

Every connector follows the same lifecycle — from authorization to an audited tool call.

AuthenticateOAuth 2.0 or scoped API key into the encrypted vault
Scope permissionsGrant the narrowest access — read-only by default
Expose as a toolTyped schema the agent can call via tool calling
Audit every callInputs, outputs & timestamps logged and replayable
Each integration moves from authorization, through scoped permissions, to an audited tool call the agent can reason over.

Credentials stay out of context

The model never sees your tokens. The platform injects credentials at call time from an encrypted vault, so a prompt injection can't exfiltrate a secret the agent was never shown.

Least privilege by default

New connectors request read-only scopes unless you explicitly grant write access. Want a sales agent to update — not just read — Salesforce? It's an intentional, logged decision.

Extend it

Build your own with the SDK

Custom tools

Any API becomes an agent tool

When the catalog doesn't cover an internal service, register it yourself. Define a name, a validated input schema, and a handler — the SDK turns it into a function the agent can call through tool calling.

Custom tools inherit the same vault-backed auth, rate limiting, and audit logging as built-in connectors, so they're production-ready from the first call.

  • Validate inputs with a typed schema (Zod or Pydantic)
  • Reuse stored credentials — no secrets in your code
  • Works with REST, GraphQL, or any HTTP endpoint
  • Automatically traced and replayable like every tool
Explore the SDKs
tools/lookup-order.tstypescript
1import { defineTool } from "@aiagents/sdk";2import { z } from "zod";34export const lookupOrder = defineTool({  // expose to the agent5  name: "lookup_order",6  description: "Fetch an order by its ID",7  input: z.object({ orderId: z.string() }),8  async run({ orderId }, { secrets }) {9    const res = await fetch(  // internal REST API10      `https://api.acme.com/orders/${orderId}`,11      { headers: { Authorization: secrets.ACME_KEY } }12    );13    return res.json();  // returned to the agent14  },15});
Registering a custom REST tool — the agent can now call lookupOrder().
By the numbers

Integration coverage at a glance

200+

Integrations

across 8 categories

8

Categories

comms, CRM, data & more

100%

Calls audited

logged & replayable

<5 min

To connect

OAuth in a few clicks

FAQ

Integrations, answered

We ship with 200+ pre-built AI agent integrations spanning communication, CRM, data warehouses, developer tools, productivity suites, and support platforms. Each connector is maintained, version-pinned, and exposed to your agents as a typed tool. If a tool isn't in the catalog yet, you can register a custom REST or GraphQL endpoint with the SDK in a few lines — so the practical ceiling is every API your agent needs to reach.

Get started

Connect your first integration today

Authorize a connector in minutes and give your agents the tools they need to do real work. Free to start — no credit card required.