Skip to content

Integrations and credentials

Integrations give session machines access to agent credentials, repositories, communication surfaces, and external tools. Configure them from Integrations in the organization dashboard.

Nori validates and stores each integration at the broker, then distributes the required access material to session machines. The exact storage, refresh, and file format depend on the integration.

The first-class Sessions agents use these credential paths:

Agent Credential options Notes
Claude Code Anthropic API key, Claude OAuth, setup token Supports interactive terminal and ACP modes
Codex OpenAI API key, Codex OAuth Supports interactive terminal and ACP modes
Cursor Agent Cursor API key ACP mode only; model selection is passed at launch
OpenCode Compatible API base URL, API key, model Installed when needed; Nori writes its agent configuration
pi-acp Compatible API base URL, API key, model Adapter and coding-agent package install when needed

Gemini credentials are not used by managed Sessions. Gemini CLI remains a separate option in the open-source Nori CLI.

For current availability, use the support matrix.

Nori supports two GitHub credential methods:

  • GitHub App: install the app and choose all repositories or a selected set.
  • Personal access token: save a token directly in the organization.

The personal access token takes precedence when both methods are configured. Repository checkout also depends on the workspace list in Fleet Setup, so a successful GitHub connection does not automatically clone every accessible repository.

Commit attribution is an organization policy. It can attribute commits to Nori, the agent, or nobody. When user attribution is enabled for a conversation, the first requester becomes the author and later participants can be added as co-authors.

Slack and Discord are session surfaces rather than generic tool credentials. Their app installations connect messages to the broker and return agent progress to the originating thread or channel.

Slack supports optional scope groups for direct messages, file upload, and file download. Configure access mode and agent reach in Fleet Setup. See the Slack guide for commands and policy.

Google Workspace uses a per-user OAuth connection. The scopes granted by that user determine which Drive, Docs, Sheets, Gmail, Calendar, and related operations their session can perform.

Nori distributes the active user access to the session rather than treating one organization-wide Google identity as every user’s credential.

The dashboard exposes a catalog of MCP servers for project management, observability, analytics, productivity, payments, design, infrastructure, databases, CI, and other tools. Authentication varies by server:

  • Dynamic client registration where supported.
  • Manually configured OAuth where required.
  • Bearer-token authentication for servers that expose it.
  • Read-only or read-write access modes for integrations that support both.

An MCP server’s own permissions still apply. Connecting it to Nori does not grant resources that the authenticated external account cannot access.

Some external command-line tools have curated credential presets. For example, the AWS preset accepts an access key, secret, and region, with us-east-1 as the default region.

Infrastructure credential distribution is disabled by default and controlled by an organization policy setting.

Custom environment variables cover services that do not have a dedicated integration. Administrators can add arbitrary key and value pairs and mark a value as admin-only so members cannot reveal it through the dashboard.

These may be long-lived secrets. Use narrowly scoped credentials and rotate them at the external service.

When an integration changes, the broker distributes the new value to running session machines. OAuth-backed integrations can also refresh access tokens and re-push them without requiring a fleet restart.

If an active agent still reports authentication failure:

  1. Confirm the integration shows as connected.
  2. Verify that the selected agent uses that integration.
  3. Check whether the user or organization account can access the requested model or external resource.
  4. Restart the agent process if the agent cached credentials internally.

For trust boundaries and secret handling, read security and credentials.

this page as markdown: /sessions/guides/integrations.md· index for agents: /llms.txt