Skip to main content

Chat Interface

Grackle's landing page is a chat interface backed by a configurable AI runtime. Type natural language commands — "connect to any codespace and start working on #454" or "what's the status of the API redesign?" — and the agent handles it using Grackle's MCP tools.

How it works

The chat interface connects a swappable AI runtime (Claude Code, Codex, Copilot, or Goose) to Grackle's MCP server. The agent sees all of Grackle's capabilities as tools — environment management, task creation, session spawning, findings, personas — and uses them to fulfill your request.

This means you don't need to memorize CLI commands or navigate the UI. Just describe what you want.

Choosing a runtime

The chat runtime is configured during the first-run setup wizard. You can change it later under Settings > Personas by editing the default persona's runtime.

RuntimeBest for
Claude CodeGeneral purpose, strong at orchestration and code tasks
CodexOpenAI model access, reasoning-heavy tasks
CopilotGitHub-integrated workflows
GooseProvider-agnostic, bring your own LLM

The chat interface uses whichever runtime your default persona specifies.

Suggested actions

On the landing page, Grackle shows contextual suggested action cards based on your current state:

  • First run — "Add your first environment", "Set up credentials"
  • No environments connected — "Connect a Docker environment", "Add an SSH host"
  • Active work — "Check on [workspace name]", "Start the next task in [workspace]"
  • Recent activity — Quick links to resume recent sessions or view findings

Click any card to pre-fill the chat with that action, or just type your own request.

What you can do

Anything the MCP server exposes is available through chat. Common patterns:

Environment management:

"Add a Docker environment called build-server and provision it"

Task workflows:

"Create a task to fix the flaky auth test in the API workspace and start it"

Status checks:

"What tasks are currently running? Any failures?"

Knowledge queries (with knowledge graph enabled):

"What do we know about the payment module architecture?"

Multi-step orchestration:

"Set up three Docker environments and start the top three priority tasks in parallel"

Ephemeral conversations

Chat conversations are not persisted across page refreshes in the current version. They're designed as a quick command interface, not a long-running conversation log. For persistent work, use tasks and findings.