connections
Everything Kadhai works with.
The models that do the thinking, the services the crew can use while it works, the steps flows can take, and the apps that can hand Kadhai work. This list comes straight from the app.
Connect in one click.
Open connect in the taskbar, pick a service and press Connect. Some need no account, some a key, some a sign-in in your browser. You choose which cooks may use each one; the Chef decides which step needs it.
tools for the crew
Connections 63
Connect a service and the cooks can use it while they work: read an issue, look up current docs, search the web, query a database. The Chef decides which step needs which connection, and you decide which cook may use it.
Up-to-date documentation for any library or framework
Ask questions about any public GitHub repository
Read the docs and code of any GitHub project
Issues, pull requests and code in your repositories
Projects, merge requests and issues on GitLab
Search models, datasets and Spaces
Search the Cloudflare developer documentation
Errors and issues from your Sentry projects
Scan code for security problems
Your Vercel projects and deployments
Search and read the official Microsoft and Azure documentation, with code samples
Search and read the AWS documentation, and which services each region has
The Svelte and SvelteKit documentation, and a checker for Svelte code
Search the Astro documentation
Logs and analytics of your Cloudflare Workers
Ping, traceroute and HTTP checks from probes around the world
Your Netlify sites and deploys
Read your Drive, Docs, Sheets, Gmail and Calendar
Pages and databases in your Notion workspace
Issues, projects and cycles in Linear
Atlassian: Jira issues and Confluence pages
Channels and messages in your Slack workspace
Tasks and projects in Asana
Tasks and docs in ClickUp
Boards and items on monday.com
Bases, tables and records in Airtable
Bug reports recorded with Jam
Customer conversations in Intercom
Files in your Dropbox
Files in your Box account
Tasks and projects in Todoist
Bookings and availability in Cal.com
Thousands of apps through your Zapier account
Search the web with DuckDuckGo — no key needed (it may refuse searches that come too fast)
Search the web and fetch pages with Exa
Web, news and image search (free tier available)
Search, extract and crawl the web (free tier available)
Turn any website into clean text
Download a web page and read it as text
Drive a real browser: open pages, click, fill forms (Playwright)
Inspect and debug a page in Chrome: console, network, performance
Read the subtitles of a YouTube video
Query and change a SQLite database file
Your Supabase projects, tables and SQL
Your Neon Postgres databases
Read and write .xlsx spreadsheets on this computer (no Excel needed)
Look at a MongoDB database: collections, schemas, queries (read-only)
Your Prisma Postgres databases
Product analytics, feature flags and errors in PostHog
Places, directions and maps through Mapbox
Turn data into charts: bar, line, pie and more
Read designs from your Figma files
Your Canva designs
Your Webflow sites and content
Your Wix sites
Customers, payments and products in Stripe
Invoices and transactions in PayPal
Payments, catalog and orders in Square
Leads, contacts and deals in Close
Contacts, companies and deals in HubSpot
Let the crew read and write files in a folder outside the workspace
A notebook the crew can save facts to and look them up in later
A scratchpad for working through a hard problem in steps
Nothing matches. Kadhai can also use any other MCP server: Connect > add your own.
the brains
Model providers 15
Add free keys from several providers and Kadhai moves to the next one whenever one says "slow down". Paid keys are optional and held to a daily token limit.
15 requests/min, 1,500/day
varies by model; for gpt-oss-120b Groq reports 1,000 requests/day and 8,000 tokens/min
10 requests/min, 100/day, 1M tokens/day
about 1 request/sec, 500K tokens/min
only models ending in ":free" cost nothing; daily request cap
up to 40 requests/min; phone verification to sign up
20 requests/min, 20/day, 200K tokens/day
a small monthly credit, then it stops
20 requests/min on a trial key
flash models free, 1 request at a time
Any OpenAI-compatible server, including a model running on this computer (Ollama, LM Studio).
plans you already pay for
Subscription CLIs 3
Signed in to a coding CLI? Kadhai can use it as a model, so your plan does the work first. On high flame it can even do a whole step as an agent, on a copy of your folder.
Use the plan you already pay for: Kadhai calls the CLI you are signed in to.
Use the plan you already pay for: Kadhai calls the CLI you are signed in to.
Use the plan you already pay for: Kadhai calls the CLI you are signed in to.
for flows
Built-in services 15
Flow steps that talk to a service directly, with no model and nothing to install. Anything else is one HTTP request away: paste a curl command or an OpenAPI description.
Send a message
Post to a channel
Post to a channel
Send a notification
Send an email
Create an issue, Comment on an issue or PR, List issues, List pull requests, Latest release, Read a file
Add a page to a database, Add text to a page, Search
Add a task, List open tasks, Complete a task
Add a card, List the cards of a list
Read the latest items
Read a subreddit
Front page, Search stories
Summary of an article, Search articles
Now and the next 3 days
Convert an amount
what starts work
Starts and triggers
Flows and tasks can start on their own. Kadhai only ever asks outward; nothing on the internet can call in.
Every few minutes, hourly, daily at a time, on chosen weekdays.
A program on this computer posts to Kadhai (local only).
A file in the folder changes.
Press Run, with pinned test data if you like.
Checked every 10 minutes.
Checked every 15 minutes.
Checked every 10 minutes.
Checked every 10 minutes.
Checked every 10 minutes.
Starts a crew task (Connect > triggers).
Starts a crew task (Connect > triggers).
Starts a crew task (Connect > triggers).
Starts a crew task (Connect > triggers).
from your editor
Apps that can use Kadhai
Kadhai is an MCP server and an ACP agent, so other tools can hand it work.
Hand Kadhai a task from a chat.
claude mcp add kadhai -- node bin/mcp-server.js
Add Kadhai as an MCP server.
Use the crew in Zed's agent panel.
Over stdio or HTTP (/mcp).