Understand
Build a map before making a move
TheGitAI searches the repository, finds relevant code, and reads the surrounding conventions before it changes anything.
- Exact, regex, and natural-language code search
- Fresh reads after files change
TheGitAI capabilities
From investigation to a verified change.
Code, browsers and desktop apps in one agent. See how it works, choose where it can act, and explore the tools that carry the work through.
Native Computer Control
Your screen, or a screen of its own.
Let it work in front of you, or keep using your computer while it works in the background. The permission prompt puts that choice in your hands, every turn or for the session.
Explore the permission choices
TheGitAI wants to use apps on this computer
Choose where it works, and whether that choice lasts for this turn or this session. Try each option below.
Background screenGNOME Wayland
Working in FreeCAD
Your mouse. Your keyboard. Still yours.
This turn
A separate screen with its own input. Keep using your own apps, or minimize the preview while the agent continues.
GNOME Wayland · Linux Mint Cinnamon · macOS on Apple Silicon and IntelInstallation and supported systems
Browser Control + Chrome extension
The browser you already use.
The session you're already in.
Work in signed-in dashboards, cloud consoles and admin panels from the same agent in your CLI or desktop app.
Get the Chrome extension Browser Control setup and permissions →Add the extension
Install it from the Chrome Web Store, then start the CLI or desktop app. It connects locally, with no pairing step.
Ask about a page
Tell the agent which page to use and what to do. Default mode asks before it acts on a website.
Stay in control
Work stays in the session’s tab group. Remove a tab from the group to take it back, or use Stop in the extension popup.
Explore the working set.
Find a capability and open its guide.
Chrome extension
Connect your existing Chrome to the CLI and desktop app.
CLI
Works from the command line next to any editor.
Desktop app
Conversations, files and changes side by side.
Browser Control
Use signed-in websites through the Chrome extension.
Computer Control
Choose your screen or the background, for one turn or the session.
MCP servers
Connect local or remote tools.
Agent skills
Teach it your workflows as SKILL.md folders.
Curated frontier models
Switch mid-session without losing the conversation.
Background jobs
Keep dev servers, watchers and local APIs running.
Parallel research
Run independent lookups together.
Live task progress
Follow the plan and its progress.
Plan mode
Investigate and plan before making changes.
Approvals
Choose when the agent asks before taking action.
Undo and checkpoint restore
Recover agent changes without overwriting your own.
Saved sessions
Resume a conversation later in the same project.
PDF, Word and Excel
Read documents and edit Word files.
Images and screenshots
Show it what you are looking at.
Web search and fetch
Look up information while it works.
Git and PR workflows
Work with branches, commits and reviews.
Steer while it works
Send corrections without stopping the run.
Verified fixes
Trace the cause, implement the fix, and prove the result against the real project.
Repository-wide changes
Keep source, tests, configuration, and documentation moving together.
Live project awareness
Keep servers, watchers, builds, and local APIs available while the agent works.
Precise recovery
Ask the agent to reverse an edit, file, turn, or checkpoint without discarding unrelated work.
One agent, the whole loop
From “where is this?” to “the tests pass.”
TheGitAI is built for work that crosses files, tools, and stages. Give it the outcome and let it gather the evidence needed to get there.
Change
Make coherent changes across files
Ask for the feature, migration, refactor, or fix. The agent can create, patch, replace, and delete files while keeping the larger task in view.
- Targeted replacements, patches, and full-file writes
- Multi-file implementation and refactoring
- Fresh-read guards against stale overwrites
Verify
Debug with the actual project
It can reproduce failures, inspect diagnostics, run focused scripts, and use your own tests, builds, linters, and typecheckers as evidence.
- Project-aware diagnostics after edits
- Shell commands and private scratch scripts
- Test, lint, build, and typecheck loops
Ship
Finish the engineering loop
TheGitAI works with Git when the task calls for it, reviews the resulting diff, and can carry work from investigation through a verified change.
- Git status, diff, branch, commit, and PR workflows
- PR-style code review and regression checks
- Clear handoff of changes and validation
Structure-aware in 15 languages
A repository map, not a bag of files
Tree-sitter-backed mapping extracts useful structure for the languages below. Other project files remain available through normal reading, listing, and search tools.
JavaScriptTypeScriptTSXPythonGoRustPHPJavaCC++C#Objective-CHTMLCSS / SCSSRuby
Momentum built in
It can keep several threads of work moving.
Long tasks should not feel like watching one command finish at a time. TheGitAI parallelizes safe research, manages persistent processes, shows visible task progress, and accepts your next instruction before the current turn ends.
Parallel tool calling
Research six independent paths at once
Reads, searches, listings, symbol queries, and web lookups can execute concurrently while edits and other mutations stay serial.
A task list you can actually see
Multi-step work stays legible with pending, active, and completed steps. The current list remains in view and the finished snapshot stays in the transcript.
Queue what comes next
Type one follow-up while the agent is busy. It submits automatically when the turn finishes, or you can recall, edit, replace, or cancel it.
Background jobs that belong to the session
Run up to eight dev servers, watchers, builds, or local APIs without blocking the conversation. Follow live output, inspect the captured log, stop a job by id, and let the session clean up what remains.
Background jobs guide● bg_1running · 4m12s
$ npm run dev
ready on http://localhost:3000
/jobs output bg_1
/jobs kill bg_1
More context than source code
Bring the spec, the screenshot, or the live web.
Real engineering work arrives in many formats. TheGitAI can bring those inputs into the same task instead of making you translate everything into a prompt first.
Documents
Work beyond plain text files
Read PDF, XLSX, and DOCX files as structured task context. Replace visible text in Word documents, including table content, and save the edited result.
- Bounded PDF pages and DOCX paragraph ranges
- Spreadsheet sheets rendered for analysis
- Validated, reviewable DOCX text replacements
Images
Show it what you are seeing
Paste a screenshot or type an image path into the prompt. Vision-capable models can inspect UI states, diagrams, error screens, and visual references.
- PNG, JPEG, GIF, and WebP input
- Clipboard paste and automatic path attachment
- Up to five images in one message (10 MB each, 20 MB total)
The web
Research without leaving the task
Search the public web and fetch specific pages as clean Markdown, with coverage and continuation metadata when the source is too large for one read.
- Search for current technical information
- Fetch public URLs and follow relevant results
- Deduplicated lookups and bounded page extraction
Past the edge of the repository
Your browser, your tools, your workflows.
Real tasks do not stop at the repository. TheGitAI works in the Chrome you are signed in to, calls the tools your MCP servers expose, and follows the skills you teach it.
Your browser
Work in the Chrome you are signed in to
With the TheGitAI Chrome extension, the agent works in the Chrome you already have open, signed in as you: cloud consoles, internal dashboards, and sites with no API or export button. It can also check the page you just built and click through the flow.
- No second browser and no separate login
- Each session works in its own tab group
- Your other tabs are never read; downloads wait for you
MCP
Plug in the tools you already use
Connect local or remote MCP servers with ai mcp add, and the model can call their tools in the middle of a task, from library documentation to browser testing.
- Local stdio and remote HTTP or SSE servers
- /mcp shows connections and available tools
- Default mode asks before each call; Plan mode blocks them
Skills
Teach it your team's workflows
A skill is a folder with a SKILL.md: the steps for a release, a deploy checklist, or a review routine, plus any scripts and references it needs. TheGitAI uses one when a request calls for it, or when you type /<name>.
- Personal skills, or shared with the repository
- Install from GitHub with /skills add
- Ask it to write a new skill for you
Power with a way back
An AI coding agent should come with a way back.
Choose the operating mode, review sensitive actions, and recover assistant changes without reaching for a destructive repository reset.
Approvals
You choose how much autonomy to give
Default mode asks before edits and commands. Bypass all permissions removes every prompt for work you already trust. Plan mode stays read-only.
- Approve once, approve the session, or deny
- Local, masked sudo password entry
- Repository containment and sensitive-path guards
Recovery
Undo the agent, not your own work
Every assistant edit is journaled. The agent can preview an undo, reverse one edit or a whole turn, and restore complete or selected files from a checkpoint.
- Conflict-aware edit undo
- Full and selective checkpoint restore
- Binary-safe recovery for edited DOCX files
Local state
Resume from the repository you were in
Recent sessions are saved on your machine and scoped to each repository, alongside a compact global prompt history for requests you reuse.
- Ten recent sessions kept per repository
- Twenty-entry prompt-only history
- Source-visible client and server-held provider keys
Local repository tools run on your machine. Relevant task context is sent to the server-backed model so it can respond and decide which tools to use. Provider credentials and private orchestration remain server-side. We don't store your source files, your prompts, or your conversations on our servers - saved sessions keep metadata only. See the privacy policy.
A frontier shortlist, not a catalog
Leading coding model families. No filler.
The picker spans complementary frontier families for speed, depth, and cost. Every entry is selected for serious coding work, and switching models does not clear the session you have already built.
Zero config: every model works with your TheGitAI account.
GPT-6 Luna (OpenAI)
GPT-6.1 Terra (OpenAI)
GPT-6.1 Sol (OpenAI)
GPT-6 Astra (OpenAI)
Claude Haiku 5.5 (Anthropic)
Claude Sonnet 5.5 (Anthropic)
Claude Opus 5.5 (Anthropic)
GLM 5.3 (Z.ai)
DeepSeek V4.1 Flash 1017 (DeepSeek)
Gemini 3.8 Flash (Google)
Gemini 4 Argon (Google)
Kimi K3 (Moonshot AI)
GLM 5.3 Flash (Z.ai)
MiniMax M3.1 Flash (MiniMax)
Grok 4.7 (xAI)
Capability index
The complete working set.
The practical features available in TheGitAI today, grouped by the job they help you finish.
Repository intelligence
- Fast whole-repository search
- Structure-aware repository map
- Exact and natural-language search
- Definitions and references
- Project diagnostics
Engineering actions
- Create and rewrite files
- Targeted replacements and patches
- Delete files safely
- Run shell commands
- Run scratch Node scripts
- Use Git and project tooling
- Call tools from MCP servers
- Follow reusable agent skills
Long-running work
- Parallel read-only tool batches
- Managed background processes
- Live and captured job output
- Visible task progress
- Queued follow-up messages
- Automatic job cleanup
Research and media
- Live Chrome browser control
- Public web search
- Web pages converted to Markdown
- PDF, XLSX, and DOCX reading
- DOCX text editing
- Image and screenshot understanding
- Clipboard and path-based attachments
Safety and recovery
- Per-action approval prompts
- Default, Bypass all permissions, and Plan modes
- Assistant edit journal
- Atomic checkpoint restore
- Stale-overwrite protection
- Secret-redacted command output
Terminal experience
- Native responsive terminal UI
- Markdown, tables, and diffs
- Mouse selection and clickable links
- Model and session pickers
- Repo-scoped saved sessions
- Browser and headless sign-in
Want the operating details rather than the tour?
Read the capability guideMissing something?
Request a feature. If it makes sense for TheGitAI, we'll build it, usually in a day or two, at no charge.
Real autonomy, still under your control
Give it the hardest task on your list.
Open a repository you know and see how far it gets on its own.
No credit cardZero configmacOS, Linux, and Windows
