Skip to main content
idapt
HomeCodeAI ModelsPricing
Sign inStart free trial
  • Home
  • Pricing
  • AI Models
  • Image models
  • Voice models
  • Video models
  • Rankings
  • New models
  • Model status
  • Multi-Model Chat
  • Agents
  • Computers
  • Drive
  • Automations
  • AI Gateway
  • All features →
  • LLM cost calculator
  • Token counter
  • All free tools →
  • Blog
  • Use cases
  • Comparisons
  • Best of
  • Benchmarks
  • Changelog
  • Help center
  • FAQ
  • Privacy
  • Compare all models
  • Support
  • idapt Code
  • Developers
  • Quickstarts
  • API reference
  • API pricing
  • CLI
  • MCP
  • Downloads
  • Desktop
  • Badges and embeds
© idapt[email protected]TermsPrivacy PolicyLegal noticeReport content
X (Twitter)
All use cases

Developers

The workspace around your editor.

Everything that isn't editing a repo: sandboxes, agents, real computers, one API.

Try nowSee pricing

What developers actually run

Set up a coding session on a paired or cloud computer: pick the machine and project, then work in scope.

Your prompt

Act as the user's coding agent on an idapt computer for this
conversation.

The `computer` playbook already carries the codebase doctrine: resolving the
machine and project, following the repository's own conventions, leaving work
you did not do alone, verifying with the project's own commands, and what to
report at the end. Do not restate it. This is the engagement contract on top.

## Scope

1. Prefer the computer and project the user named. Ask for the path or clone
   URL when it is missing rather than guessing, and ask which machine only
   when more than one is plausible.
2. If no computer is available, send the user to the existing Add Computer
   flow. Do not invent a machine or silently substitute a sandbox.
3. Work inside the scope you were given. Ask first before a material expansion,
   a system-wide install, or anything that touches production.
4. Prefer focused, reversible changes and dependencies the project already has
   over introducing new ones.
5. Keep work inside the selected project unless the task requires otherwise.
   Shell access can still reach anything the daemon's operating-system user
   can reach, so scope is a commitment you keep, not a wall you are behind.

## Credentials

Refer to workspace credentials by NAME and let the server inject them. If one
is missing, point the user at the workspace Credentials page. Never ask them to
paste a credential into chat, and never print credential files or dump the
environment. Prefer narrowly scoped, repository-only tokens, and use a write
credential only for the operation that needs it.

## Reporting

Say plainly what you did not verify. A change whose tests you did not run is
unverified work, and naming that is worth more than a confident summary.
Use idapt CodeOpens with everything above already filled in.

Your editor's AI stops at the repo boundary.

idapt owns everything around it: research, sandboxed runs, real machines, automation.

Spinning up a scratch VM for one experiment is a chore.

A cloud computer is one message away, and the agent can drive it for you.

1

Pick the right model per task

Cheap model for boilerplate, frontier model for the hard call: same chat.

More about Multi-Model Chat
2

Run it for real

Sandboxed Python, Node, and shell against your actual Drive files.

More about Code Execution
3

Give agents a machine

Long builds, real databases, real dependencies: on paired or cloud computers.

More about Computers

Everything developers can run in one click

Code

  • idapt Code
  • Fix a failing test
  • Fix a GitHub issue
  • Find security bugs
  • Ship a feature

What this replaces

a chat tab beside your editora separate scratch cloud VMper-provider API keys and SDKs

Built from these features

Code Execution

Sandboxed Python, Node, and shell: agents and you, side-by-side.

Computers

Daemon-connected machines with real filesystems and agent-friendly controls.

Cloud Computers

Spin up cloud machines your agents can use: per-second billing, torn down when idle.

Agents

Skills, scoped permissions, persistent memory: any model you pick.

AI Gateway

OpenAI-, Anthropic-, and OpenRouter-compatible endpoints with routing you control.

CLI

Script files, agents, chats, and 200+ models: from your shell.

Frequently asked

Ready to work this way?

Try 200+ models free for 7 days, then pick the plan that fits. Cancel anytime.

Try now

Every provider means another key, SDK, and rate limit.

One key and one OpenAI-compatible endpoint reach 200+ models.

4

Wire it into CI

The typed SDK and versioned API drive all of it from code.

More about API & SDK
  • Speed up slow code
  • Untangle a god function
  • Migrate to strict TypeScript
  • Upgrade a dependency
  • Resolve a merge conflict
  • Fix a broken build
  • Raise test coverage
  • Explain a codebase
  • Review a change
  • Write tests
  • Build a web app
  • Write a SQL query
  • Clone a repository
  • Review a branch
  • Open a pull request
  • Triage issues
  • Find a regression
  • Upgrade dependencies
  • Run the tests
  • Write a bug report
  • Write a postmortem
  • Review for accessibility
  • Write a commit message
  • Design a database schema
  • Build a portfolio site
  • Recreate a page you like
  • Build a business site
  • Build an MCP server
  • Write an eval suite
  • Build a quote calculator
  • Build an event page
  • Build a pricing calculator
  • Build a CLI tool
  • Document an API
  • Scaffold a project
  • Build a dev utility
  • Build a landing page
  • Publish a style guide
  • Build a before and after page
  • Build a size guide
  • Build a self-assessment
  • Build a lead quiz
  • Build a clickable prototype
  • Write a postmortem
  • Assess a legacy codebase
  • Audit your dependencies
  • Threat model a system
  • Tune a slow query
  • Document an architecture
  • Write on-call runbooks
  • Audit your infrastructure code
  • Map tests to requirements
  • Build a mock API
  • Audit your Kubernetes manifests
  • Port legacy code
  • Build an internal admin tool
  • Write end-to-end tests
  • Audit page performance
  • Find your tech debt
  • Build a load test
  • Find your flaky tests
  • Generate a typed SDK
  • Write a migration guide
  • Find leaked secrets
  • Harden your CI pipeline
  • Slim a Docker image
  • Make an app translatable
  • Triage authentication logs
  • Write property-based tests
  • Build a commission calculator
  • Convert a spreadsheet macro
  • Migrate a database dialect
  • Build a staff directory

    • Computers

    • Run this on a computer
    • Spin up a computer
    • Set up a dev environment
    • Run a container
    • Run a dev container
    • Deploy a Compose stack
    • Expose a port
    • Back up a container
    • Move a container
    • Set up a dev environment
    • Browse a site and extract
    • Deploy a function
    • Fill a form from your details

    has a recorded run. See every skill