Public Agents#

A public agent on idagents.ai is an AI agent you create through the web dashboard and deploy to the shared EU runtime. Once deployed, the agent gets a public URL like https://idagents.ai/a/<slug> that anyone can talk to over chat, REST-AP, or MCP.

This page covers the SaaS flow at idagents.ai. For the open-source CLI flow, see Quick Start.

Sign-in is Google-only. Visit www.idagents.ai/signin and click "Sign in with Google". Your account is created automatically the first time you sign in. Use the same Google account every time — accounts cannot be merged.

Beta limits#

ID Agents on idagents.ai is in beta. Important constraints:

One agent per account. During beta you can have at most one public agent at a time. Delete it from the Danger zone to create a different one.
Free. Billing is off during beta. Plans and limits may change at any time.
No SLA. Agents may stop, restart, lose state, or be paused. Knowledge documents and deployment history may be lost. Keep your own backups.
See the Terms of Service for the full picture.

Create an agent#

From the dashboard, click Create agent and fill in:

Slug — lowercase letters, numbers, single hyphens. Becomes the public URL: https://idagents.ai/a/<slug>. Cannot be changed after deployment without recreating.
Display name — what shows in the dashboard and on the public agent page.
Role — short description of what the agent does (optional).
Description — longer prompt-style text that shapes how the agent responds. Edit any time.
Tier — Shared is the default. Dedicated is not yet enabled.

Saving the form creates a draft. Drafts are not yet running and have no public URL.

Deploy#

From the agent page, click Deploy. The platform allocates a slot on the shared EU runtime, installs a Juno instance under the hood, and brings the agent online. This typically takes a minute or two.

Once status flips to deployed, the agent has a working public URL and a green Live badge.

The public URL surface#

Every deployed agent exposes three endpoints under https://idagents.ai/a/<slug>:

Path	Method	Purpose
`/.well-known/restap.json`	GET	REST-AP discovery — capabilities, endpoints, agent identity
`/mcp`	POST	Model Context Protocol — tools/list, tools/call
`/talk`	POST	Chat — send `{"message":"…"}`, receive a reply

The agent detail page in the dashboard shows the three URLs as a vertical list with copy buttons. Share any of them with clients, embed the chat URL in a frontend, or point an MCP-aware tool at /mcp.

There is also a /health endpoint that returns the agent's runtime status.

Remote MCP#

The /mcp endpoint on every public agent is a remote MCP server. It speaks the Model Context Protocol over HTTP, which means any MCP client that supports remote (HTTP-transport) servers can connect to it without you installing anything locally.

The endpoint URL is:

https://idagents.ai/a/<your-slug>/mcp

For example, you can add the URL to Claude.ai as a custom MCP server and start chatting with your agent right from the web app. Other AI chatbots that support remote MCP work the same way.

Because the endpoint is public, anyone with the URL can connect their AI client to your agent. There is no per-client auth. If your agent exposes sensitive tools, do not give the URL out broadly.

To see what tools your agent exposes, hit the MCP tools/list button in Live controls or POST {"jsonrpc":"2.0","method":"tools/list","id":1} to the URL directly.

Knowledge documents#

You can upload markdown files as knowledge documents from the agent page. Files are stored privately, sanitized server-side, and made available to the agent at runtime.

After upload, click Sync knowledge to agent to push the latest knowledge onto the running instance and restart it.

Live controls#

The dashboard's Live controls panel lets you smoke-test a deployed agent without going through the public URL. Buttons:

Check now — runs a one-off health probe and updates the live status indicator.
REST-AP — fetches /.well-known/restap.json from the agent and shows the response body and headers.
MCP tools/list — POSTs a tools/list JSON-RPC call to /mcp and shows the response.
Chat — sends a message to /talk from the dashboard and shows the reply.

These calls go through the manager (not the public URL), so they bypass per-tier rate limits and always reach the agent directly.

Logs#

The View logs button on the agent page opens a paginated health-check history with newest entries first. Use More at the bottom to load older entries. History is polled every 15 seconds while the agent page is open in any tab.

Pausing and deleting#

Two destructive actions live in the Danger zone at the bottom of the agent page, both two-step confirmation:

Pause — stops the agent. The deployment record stays; you can resume later with the Resume button. Knowledge and settings are preserved.
Delete deployment — tears down the running instance and schedules the agent and its data for permanent deletion. Cannot be undone.
Delete agent — permanently removes the agent, its knowledge documents, and deployment history. Cannot be undone.

What gets stored#

When you create and deploy an agent we store:

Your Google email address (for sign-in).
The agent profile you typed (slug, display name, role, description, tier).
Knowledge documents you upload (markdown content, sanitized).
Deployment job logs and health-check history.
Provider key usage counters (per-agent OpenRouter spend caps).

We do not store the contents of conversations between visitors and your public agent beyond what the agent itself decides to retain.

What's onchain#

Public agents created through the dashboard are not automatically registered onchain. The two reference public agents we run (docs.idagents.ai, help.idagents.ai) are registered on Base via ERC-8004 and ID Chain ENS, but that flow is operator-only at the moment.

If you want your agent registered onchain, contact @idchain_world on X.

Troubleshooting#

Deploy says "Last deploy attempt failed" — open the agent page; the error from the failed job is shown verbatim. Most failures are transient. Click Deploy to retry.
/talk hangs — the agent process may have hit a tool-use prompt and blocked. Use Pause then Resume to restart it cleanly.
Health controls show Down — the agent's REST-AP server isn't responding. Pause and resume; if it persists, delete and redeploy.
Cannot create a second agent — beta limit. Delete the existing one first.