Registry integrations
The registry provides prebuilt agent integrations for popular platforms. You can copy a registry entry into your project, supply your own credentials, and have a working agent without writing the client yourself.
Registry entries are copied by value into your project. The agent code and its secret requests are copied; secret values are not copied. You supply those yourself after copying.
Use a registry integration
- Open Agents and browse the registry. Without a project selected the page is titled Agent Registry; from inside a project it is titled Copy Agent.
- Each entry lists its Agent name, the publishing Organization, its Scope, and the number of Required secrets.
- Select an entry to open its detail page. There you can read the description, see the Required secrets it needs, and review a read-only Code preview.
- Select Copy from registry.
- On the copy form, choose the Destination project (if you are not already in one) and set an Agent title. The required secret keys are listed for reference.
- Confirm with Copy from registry. The agent is created in your project with its secret requests in place.
After copying, open the agent and create the matching project secrets. The agent detail page flags any Missing secrets and links you to create them. Then preview the agent to confirm it works before using it in a specification.
Available integrations
Each integration has its own page covering what it connects to, the secrets it needs and where to get them, its turn support, and any limitations.
LLM and text generation
| Integration | What it does |
|---|---|
| ChatKit Workflow Agent | Run an OpenAI ChatKit workflow and return assistant text. |
| Databricks Single-Turn Responses Agent | Call a Databricks ResponsesAgent without retaining conversation history. |
| Databricks Client-History Responses Agent | Keep complete Responses item history in the Spec27 sandbox session. |
Sentiment
| Integration | What it does |
|---|---|
| Comprehend Sentiment Agent | Return sentiment from Amazon Comprehend. |
| NLP Sentiment Agent | Return sentiment from the Google Cloud Natural Language API. |
Chat platforms
| Integration | What it does |
|---|---|
| LangGraph Hosted Assistant Agent | Call a hosted LangGraph assistant and return the latest reply. |
| Botpress Chat API Agent | Hold a Botpress Chat API conversation and return the next bot reply. |
| Decagon Chat Widget Agent | Hold an anonymous Decagon web chat conversation and return the next plain-text reply. |
| Intercom Fin AI Agent | Ask Intercom Fin and return the AI Agent reply. |
| Zendesk Advanced AI Agent | Hold one Zendesk Advanced AI Agent conversation and return the whole business reply sequence. |
| Zendesk Essential AI Agent (Legacy) | Hold a Zendesk Web SDK conversation on the legacy Answer Bot path and return a business reply. |
| Freshdesk Web Chat Agent | Hold a Freshdesk Web Chat conversation with an AI Agent as a visitor. |
| Gumloop Agent Session | Hold a multi-turn conversation with a Gumloop agent through its session API. |
| Jotform AI Agent Chat | Hold a persistent Jotform AI Agent conversation and return the next reply. |
| OpenDialog Single-Turn Agent | Send a single message to an OpenDialog scenario. |
| OpenDialog Multi-Turn Agent | Hold a multi-turn OpenDialog conversation. |
Some integrations preserve conversation state across turns, which makes them suitable for multi-turn evaluation. Their pages note this under Turn support. To write your own multi-turn agent, see Writing a multi-turn agent.
Publishing your own
An organization can publish a reusable agent into its own registry so teammates can copy it into their projects, the same way they copy a built-in integration.