Autopilot and chat

AgentRuntime provides two chat surfaces:

• Console Chat (/chat) — Autopilot copilot with thread list and streaming responses
• Workspace chat (space.agentruntime.io) — Team channels and agent rooms (separate app)

This page covers Console Autopilot and how it relates to workflows.

​

Autopilot modes

Autopilot is the AI assistant built into the Console. It streams responses in the Chat panel and can take actions on your behalf: Autopilot uses your project context (tenant, project, role) and calls BFF endpoints at /v1/autopilot/*.

​

What Autopilot can do with workflows

Today Autopilot supports workflow-oriented actions including:

• create_workflow — Start a new workflow from a description
• scaffold_starter — Generate a starter graph template
• run_workflow — Start a run on an existing workflow
• Navigate — Open Workflow Studio or Command Center

Ask Autopilot to “create a workflow that emails a summary” or “run the daily report workflow” to use these actions.

​

Chat threads

The Chat page lists conversation threads persisted per user and project. Messages stream over /v1/autopilot/stream with token-by-token delivery. Threads are separate from workflow runs, but messages can reference runs, workflows, and pending human tasks as the integration matures.

​

Platform agents

Agents (sidebar, feature-flagged in some environments) are reusable agent definitions installed in your project. They participate in dm and channel rooms through the agent service — distinct from Autopilot. Platform agents run multi-turn loops with MCP tools and memory, orchestrated by agent-service rather than the workflow DAG engine.

​

Workspace chat (space)

The team workspace at space.agentruntime.io provides Slack-shaped channels, Autopilot DM, threads, and workflow system cards. It uses the chat-service API and shares authentication with the Console via .agentruntime.io cookies. Use Console Chat for operator copilot tasks; use workspace chat for team coordination and agent room participation.

​

Memory (knowledge layer)

AgentRuntime includes a memory layer for long-term knowledge — search, extraction, and episodic memory scoped by tenant and project. Memory tools are available to agents and workflow prepare-step hooks. Console memory UI may be feature-flagged; the APIs are available at /v1/memory/* for contributors.