Connections
Connections is where you give your agents access to the platforms you use. Open Settings → Connections to see every connected platform, its health and which projects can use it. When an agent needs access it does not have, it suggests the connection right in the chat.
Real Axis UI with deliberately seeded, privacy-safe demonstration content.
Platforms people connect
Section titled “Platforms people connect”Any platform with an API, a command-line tool or an MCP server can become a connection. Platforms Axis users connect today include:
| Area | Platforms |
|---|---|
| AI and models | OpenAI, xAI, ElevenLabs, Anthropic (Claude) |
| Code and hosting | GitHub, Railway, Render, Supabase, RunPod, Amazon Web Services, Google Cloud |
| Work and docs | Google Workspace, Gmail, Notion, Slack, Figma, Trainual |
| Business | Stripe, PostHog, Twilio, Namecheap, X |
| Data | Supadata |
Model providers you chat with are set up separately, in the model picker. See Model providers.
How an agent suggests a connection
Section titled “How an agent suggests a connection”When a task needs a platform you have not connected, the agent first works out how access works for that platform: its official API or MCP server and exactly which credentials are needed and where to find them. It then shows a card with the platform’s logo, the reason, and a Connect button. The connect window puts the recommended method first and lists each field you need.
Platforms you have already connected, or dismissed, are not suggested again.
Where your credentials live
Section titled “Where your credentials live”- By default, secrets are encrypted (AES-256-GCM) in a vault on your own server.
- You can point a field at an existing
.envfile, a credentials file, an environment variable, or a 1Password reference instead. - Secret values never appear in connection records, chat transcripts, logs or tool output. When an agent runs a command with a connection, the values are passed in as environment variables and redacted from anything that comes back.
Scope each connection
Section titled “Scope each connection”A connection is available either to all projects or to a list of projects you choose. One created from inside a project defaults to that project only. Agents can only use connections that are in scope for the project they are working in.
Health
Section titled “Health”Testing a connection runs a real check: for an MCP server it lists the server’s tools, otherwise it makes the platform’s verification request. Only a passing check shows a green Connected badge. A failed check shows Fix, which edits the same connection rather than creating a second one.
Good habits
Section titled “Good habits”- Use the narrowest credential the platform offers, such as a read-only or project-scoped key.
- Agents still ask before consequential actions on a connected platform: sending, publishing, payments, deletions.
- Never paste secrets into a chat, a Doc or
AGENTS.md. Use the connect window.