Allowed
Runs right away under the agent’s grant.
Product·Build a connector
Build a connector for your product
Give agents a way to use your product — CRM, docs, billing, or an internal tool. Wire any application to any agent in Neuro OS. Each company decides who can use the connection.
Any appConnect the products your agents already need
Any agentGrant access per role in the org chart
GovernedAllowed, Ask first, or Off — per action
YoursScopes, secrets, and identity stay under company control
Choose how it works
Three ways your product talks to agents
A connector is how Neuro OS reaches your product — and how people reach agents from inside it. Start with the shape that matches one real customer workflow.
- Agent tool — Agents call your product’s actions while they work: create a deal, fetch a doc, post a message.
- Chat — People message agents from inside your product, on the same rules as Telegram or Slack.
- Model provider — Agents use your models through an AI connection in supported runtimes.
Permissions
Each company decides who can use the connection
Connectors inherit Neuro OS governance. You expose the actions; the company sets who may run them, under which identity, and with which scopes.
- Tool actions — Allowed (runs now), Ask first (waits in Review), or Off.
- Agent access — Companies choose which agents can use the connection.
- Account identity — Act for the company, one person, or one dedicated agent.
- Provider access — Request only the documented scopes your actions need. Neuro OS does not expand an issued key’s permissions.
How it works
From one workflow to a live connector
Add your product name, a concrete customer workflow, and developer docs. Ship the smallest useful surface first — then expand actions once the path is proven.
Start with one workflow
Name the product, the job an agent should do, and the official docs. Prefer one read path and one governed write over a kitchen-sink API dump.
Build with the docs
Follow the Neuro OS connector guide: actions, auth, scopes, identity, and audit. Reuse official remote servers where they exist.
Review and ship
Prove discovery, an allowed read, a governed write where authorized, denial, reconnect, and revocation. Then publish to the companies that need it.
Starter prompt
Start with one useful customer workflow
Paste this into your coding agent. Replace the brackets, then let it draft a bounded plan before any implementation.
Connector starter prompt
We are the team behind [PRODUCT NAME]. Help us prepare a Neuro OS connector for this customer workflow: [CUSTOMER WORKFLOW]. Official developer documentation: [DOCS URL]. Provider server or API documentation, if separate: [SERVER OR API DOCS URL]. First read the current Neuro OS docs: https://www.patternautomation.com/pattern-automation/help/neuro-os-docs/ Search for an existing connector and related work. Preserve unrelated changes. Start with a bounded plan: the customer workflow, proposed actions, supported connection method, required account access, missing prerequisites and acceptance checks. Do not invent capabilities or assume writes start disabled. Use our official provider documentation. Verify authentication, scopes, account or plan requirements, limits and action risks. Keep credentials in approved secret references. Preserve Neuro OS access controls, action policies and audit boundaries. Document setup, supported actions, account access, restrictions, revocation and maintenance. Use a concrete action-based tagline. Derive any safety warning from this provider’s own documentation. Run the smallest relevant deterministic tests first. Ask us for test-account authorization before live proof and before any live write. Check setup, tool discovery, an allowed read, a governed write where supported and authorized, a denied call when the method declares one, reconnect, revocation and audit evidence. Keep secrets out of logs. Deliver the implementation, setup documentation, test evidence and maintenance notes. Do not open a PR, merge, deploy, publish or perform risky actions without our authorization. Begin with the plan and questions that block it.
“Any application can reach any agent — as long as the company still decides who may act, under which identity, and with which scopes.”
Keep exploring
Product · Extensions
Skills & Extensions
Skills, plugins, and MCP tools layered on top of your agents.
Learn more →Product · Channels
Channels
Telegram, WhatsApp, Slack — where people already talk to agents.
Learn more →Product · Agents
Bring Your Own Agent
Adapters for Claude, Codex, Hermes — on your own keys.
Learn more →Connect your product to Neuro OS
Ship a connector so any agent can use your app — under company governance.