Your first session

A session is one running conversation with an AI coding agent inside a working directory. Glueprint shows what the agent does in real time, asks you to approve risky actions, and saves the whole transcript.

Pick an agent

Glueprint launches an external coding agent on your machine; it doesn’t ship its own model. There are three first-party agents and a growing catalog of community agents reached through the Agent Client Protocol.

First-party agents — detected automatically when you install them:

• Claude Code — from Anthropic.
• Codex — from OpenAI.
• Gemini CLI — from Google.

ACP agents — any agent that speaks the Agent Client Protocol is also supported. Browse the ACP agent registry to see what’s available, install one per its docs, then point Glueprint at the binary in Settings > Agents.

Open Settings > Agents to see which agents Glueprint has detected. If an agent is missing, install its CLI and relaunch the app — detection runs at launch. The order shown in Agents is the order Glueprint suggests when you start a session.

Pick a working directory

A session is always tied to one folder — usually a git repository. Multiple sessions can share the same folder; Glueprint deduplicates the heavy lifting so file watchers and git status calls don’t multiply.

If you haven’t picked any folders yet, use Add a project on the setup checklist, or open Settings > Projects. The folder picker is also offered the first time you start a session.

Start a session

Glueprint opens on Workspaces, where the setup checklist walks you through Git, an agent, a project, validating and logging in to the agent, and then your first session. No account is needed for any of it.

1. Click Start a session on the setup checklist, or New session in the sidebar.
2. In the composer, pick the agent and the project. Keep the project folder, or choose New worktree to give the session its own branch.
3. Write your first message and click Send.

The session opens as a tab in its workspace, with a pulsing status dot in the sidebar while the agent works. The tab shows messages, tool calls, file edits, and command output as they happen.

You can start more than one session at the same time. Each session runs in its own subprocess and has its own conversation history.

Approve actions

When the agent wants to do something risky — write a file outside the working directory, run a shell command, talk to a non-allowlisted tool — Glueprint pauses and asks. A pinned card appears at the top of the session with the requested action and Allow / Deny buttons.

Each session has a permission mode, such as Ask, Accept Edits, or Plan; press Shift+Tab in the chat input to cycle it. Approval cards can also always allow a specific tool right from the chat. Defaults are intentionally cautious; loosen them as you build trust with a workflow.

Pause, resume, stop

• Pause keeps the session alive but tells the agent to wait. Useful when you want to read what’s happened so far without new output piling on.
• Stop ends the session. The transcript stays in the sidebar; you can open it any time and start a follow-up session from where it left off.
• Reopening a stopped session and sending a new message resumes the same conversation when the agent supports it.

See your session from another device

On Pro and above, once this desktop is connected to your account, the web portal at portal.glueprint.ai shows your sessions live, and you can answer their prompts there. Below Pro, the portal manages your account and hosts; see Plans. Transcripts, prompts and approvals are end-to-end encrypted between your machine and the other surface; the relay routes them but never sees their content.

Next step

Continue to the Glossary to learn the rest of Glueprint’s vocabulary, or skip ahead to Sessions for the full reference.

Prefer the keyboard? Cmd+K (Ctrl+K on Windows and Linux) opens the command palette, Cmd+1 to Cmd+9 switch workspaces, and Cmd+/ shows every shortcut. See Keyboard shortcuts.