Skip to main content
Tascarrel is alpha software and may break. We’re looking for your feedback — share it on GitHub.
Documentation

Run Your First Task

This guide creates a workspace through the UI, configures one coding harness, and runs one reviewed agent task.

Start Tascarrel

The server installer starts the per-user service. If you stopped it after installation, start it again with:

tascarrelctl daemon start

Create a key for the first browser:

tascarrelctl auth pair --label "Local browser"

Open the UI at http://tascarrel.localhost:8272 and enter the printed key. Tascarrel retains the resulting browser session. If the UI does not load, check:

tascarrelctl daemon status
tascarrelctl daemon logs

Run tascarrel directly to keep the server in the foreground.

If you installed the experimental desktop application, open Tascarrel from the macOS or Linux application launcher instead. Its startup window reports host checks and payload extraction, pairs its browser session through the private control socket, and loads the application when the server is ready.

Create and Start a Workspace

Use the workspace creation page in the UI:

  1. Select Create workspace.
  2. Enter demo as the workspace name.
  3. Add the Git repositories for the first task and select their branches when needed.
  4. Select the languages, tools, and workspace capabilities the project needs. Leave optional settings at their defaults when unsure.
  5. Review the generated Dockerfile, config.toml, and AGENTS.md, then select Create workspace.

Tascarrel creates the workspace and starts its virtual machine. The first start may take a moment while the guest becomes ready.

Configure a Coding Harness

Before starting a task, open Settings in the workspace sidebar and configure at least one coding harness:

  • For Codex, open Settings → Harnesses, select Install harness, and then select Sign in with ChatGPT. Complete the displayed device-code flow.
  • For Claude Code, open Settings → Harnesses, select Install harness, run claude setup-token separately, and enter the resulting setup token.
  • For Tasci, open Settings → Tasci, add an OpenAI-compatible endpoint and model, and select a default model. Tasci is bundled with Tascarrel.

Harness credentials are shared by every pod in the workspace. Use separate workspaces when projects should not share agent credentials.

Start a Task

Use the add button beside Pods and enter a concrete first request, such as:

Inspect the repositories in this workspace and summarize what each one does.
Do not change any files.

Tascarrel creates a pod and its first chat together. The initial image may take time to build, so the first pod usually starts more slowly than later pods.

While the agent runs, you may leave the page. Tascarrel keeps the chat and shows an attention indicator when input, a failure, or a completed turn needs review.

Find Your Way Around

The sidebar selects workspaces and pods. Workspace-level entries manage repositories, images, network access, and settings. A selected pod provides:

  • Agent for chats and prompts.
  • Code for the embedded code editor.
  • Changes for Git status, commits, and diffs.
  • Files for quick previews below /workspace.
  • Pod for lifecycle details, supervised processes, and repository imports.

Capabilities such as USB appear only when enabled.

When you are finished, stop the pod to retain its state or destroy it to remove its private files and processes.

Stop without Deleting

Stop the workspace virtual machine:

tascarrelctl workspace stop demo

Stopping retains the workspace configuration, persistent disk, images, pods, chats, and repository state. Starting the workspace again restores the retained resources, although processes that were running before the VM stopped return as stopped records rather than live processes.

Next, choose the right boundary or configure the workspace.