An agent that finishes the job
It keeps going until the task is done, and shows its work as it goes.
- Live to-do list for every task
- Subagents for focused side work
- Web search and research with sources
- Semantic search across your code
An AI engineer that works in your codebase
Describe the change you want. Kai Agent plans it, writes the code, runs it in a private sandbox and opens the pull request. You stay in charge of what ships.
$1 of AI usage every month. No card required.
- Find where sign-in is handled
- Add a limit on repeated attempts
- Cover it with a test and run the suite
read_filesrc/auth/auth.controller.ts
write_filesrc/auth/auth.controller.ts
@Post('login')+ @Throttle({ default: { limit: 5, ttl: 60_000 } })login(@Body() body: LoginInput) {
bashpnpm test auth
All tests passed
Ready to open a pull request
From request to pull request
You describe the outcome. The agent does the work where you can watch it, stop it and undo it.
- Step 1
Describe the change
Ask in plain language, attach a screenshot, or point at files with an @-mention.
- Step 2
The agent makes a plan
It reads your code and lists its steps. You can have it stop at the plan until you agree.
- Step 3
It works in a sandbox
It edits files, runs commands and tests, and fixes what breaks, all in a private workspace.
- Step 4
You review and ship
Read the diff, roll back anything you dislike, and open the pull request.
Everything the job takes
More than a chat box: a workspace, tools and guardrails, all in your browser.
A private sandbox for every chat
A real Linux workspace, so the agent runs what it writes instead of guessing.
- Files, terminal and a browser
- Live preview of the app it builds
- Your files stay put when it goes idle
- No route to private networks
You decide what it may do
Choose how much freedom it has, chat by chat, with rules for your whole team.
- Manual, auto and plan-only modes
- Approve risky actions as they come up
- Allow and deny rules for tools and commands
- Hooks that can block a tool call
Nothing is permanent
Every edit is reviewable, and every step can be undone.
- Side-by-side diffs of each change
- Automatic checkpoints before edits
- Restore any checkpoint in one step
- Fork a chat to try another approach
Built around GitHub
Connect a repository and the agent works the way your team already does.
- Works on a branch and commits there
- Opens the pull request for you
- Automatic pull request review
- Mention it in a comment to get help
Teach it how you work
Give it your conventions once, and extend it with your own tools.
- Project rules and custom commands
- Skills and plugins
- Connect remote MCP servers
- Memory that carries across chats
Made for teams
Share projects, keep spending in check, and see what happened.
- Organizations, projects and roles
- Discussion Board with your agents
- Scheduled and background tasks
- Spend limits and usage tracking
Answers you can use
Not everything is a code change. Ask for what you actually need.
- Explain unfamiliar code
- Attach images and screenshots
- Charts and files you can keep
- Compact long chats and keep the context
A boundary around everything the agent runs
An agent that runs commands has to be contained, so Kai Agent keeps it apart from your systems.
Sandboxes are kept apart
Each organization has its own sandbox network, and sandboxes run on separate machines from the service that holds your credentials.
No way into private networks
Code in a sandbox can reach the public internet, but it cannot open connections to private addresses or cloud metadata.
Credentials are encrypted
Stored secrets are encrypted at rest, and every request is checked against your access to that project.
Try it free, then pick a plan
Every new workspace can send 2 test messages free. Each plan then gives you a monthly allowance of AI usage, and you can add more from Settings whenever you need it.
Questions
Can I try it before paying?
Yes. Every new workspace can send 2 test messages on the low-cost models, with no card required. After that, choose a plan in Settings to keep going.
How is usage counted?
AI usage is metered at the model price, so a small task costs a few cents and a long agent run costs more. You can see your spending in the app, and the agent stops when you reach your limit.
Which AI model does it use?
Kai Agent runs on GPT-5, GPT-5 mini and GPT-5 nano, and on Claude Haiku 5.5 and Sonnet 5.5, all hosted on Azure AI Foundry. Starter uses the low-cost models; Pro and Team can use every model.
What happens to my sandbox when I stop working?
After 10 idle minutes it stops and keeps your files, and it wakes up when you send the next message. That can take a few minutes. A workspace is deleted 3 days after it was created, so push anything you want to keep to GitHub.
Can it change my code without asking?
Only if you let it. In manual mode you approve risky actions one by one, plan mode makes no changes at all, and auto mode lets it work freely inside the sandbox.
How do I pay for a plan?
Choose a plan in Settings once you are signed in, and pay in Indian rupees by UPI, card, net banking or wallet through Razorpay. Plans renew only when you pay again, so nothing is charged automatically. See the pricing, refund and terms pages linked below.
Give it a real task
Create an account, confirm your email and start your first chat.
Create your account