AI Agents

AI agents are the core of Unified Dev. You can delegate any issue or task to an agent, watch it work step-by-step in real time, review the generated diff, and merge — all without touching a browser or switching terminals.

Supported Providers

Unified Dev works with the following AI providers out of the box:

• Claude Code (Anthropic) — set ANTHROPIC_API_KEY in your environment or in Settings → Agents
• OpenAI Codex — set OPENAI_API_KEY
• GitHub Copilot CLI — sign in via Settings → Integrations

You can run multiple agents in parallel across different repositories or threads. Each thread uses the model you select for that thread independently.

Plan Limits

• Free — 10 agent runs per day, 3 active threads, 3 repositories, 1 organization, local only
• Pro — Unlimited agent runs, unlimited active threads
• Ultimate — Everything in Pro, plus remote access and mobile access

See Pricing or License Activation to upgrade.

Creating a Thread

A thread is a persistent conversation between you and an agent, scoped to a single repository. Each thread lives in its own git worktree so it never interferes with your working tree.

• Open the Agents section from the sidebar
• Select a repository from the agent sidebar
• Click Add new thread (or use the New thread keyboard shortcut)
• Optionally, create a thread from an existing issue, pull request, or branch using the source picker

Sending Messages

Type your task in the chat input and press Enter. You can use / to open the slash command menu, which lists built-in commands and installed skills.

Slash commands

• /plan — switch to plan mode (the agent drafts a plan before executing)
• /think — enable extended thinking budget for complex tasks
• /fast — disable thinking for quick, low-latency responses
• Any installed skill name — invokes the skill's prompt

Attachments

You can attach images to any message. Use the attachment button in the chat input or drag and drop an image file. Images are included in the context sent to the AI provider.

Monitoring Progress

Once the agent starts, the timeline shows each step as it happens:

• Each tool call (file read, shell command, search) appears as a timeline entry
• The status bar shows a live count of tool calls made
• You can stop the agent at any time using the Stop agent button in the header

Reviewing the Diff

When the agent finishes, the diff viewer opens automatically on the right side of the workspace. It shows all file changes, grouped by file. You can:

• Browse changes file by file
• Edit any file directly inside the app using the inline file editor
• Discard individual file changes with the discard button
• Toggle all changes at once

Merging and Creating PRs

When you're happy with the diff, use the action button in the agent header to choose what to do next:

• Merge locally — merges the agent's branch into your base branch on your machine
• Merge and push — merges locally and pushes to the remote
• Draft PR — creates a draft pull request on GitHub
• Create pull request — creates a published PR directly on GitHub
• Push changes — commits and pushes to update an existing PR

The default merge action can be configured in Settings → Behaviour → Default Merge Action.

Integrated Terminal

Each thread has an integrated terminal panel that runs inside the worktree directory. You can open multiple tabs, run tests, inspect the file system, or execute any shell command. Toggle it with the keyboard shortcut shown in Settings → Shortcuts → Toggle Terminal.

File Explorer

The file explorer panel lets you browse all files in the worktree. Use the search box at the top to filter by filename. Click any file to open it in the inline editor.

Skills

Skills are reusable prompt templates that extend agent capabilities. You can install skills from the Skills section in the sidebar and invoke them via /skill-name in any thread.

MCP Servers

Model Context Protocol (MCP) servers give agents access to external tools such as GitHub, Jira, Notion, Sentry, and Slack. Manage connected MCP servers via the MCP section in the agent sidebar.