Free local and cloud AI
One conversation for browser tasks, plans and Agent Teams.
Start a conversationInteractive preview
How it works
One conversation. The right tools for the job.

A simple page for a first inspection.
More information…No changes made.
Illustrated sample. No live task runs here.
Apps
Open an app when you want direct control. It opens as a window in your Room, right next to the conversation.
Choose an app to see its window.

Agent Teams
Say what needs doing. Codelit finds the proven Agent Team for it and shows exactly what it hands back.
Matched onshippreview
Or start from a proven outcome
How a Team starts
Best match: Release Safety Team.

Turn a scoped issue into a branch, preview, browser check, pull request, and release proof.
Works with
You get back
Approvals and receipts
Approval gates pause before a real write. Run progress, evidence and the receipt stay in your Thread, in order.

Includes the estimated cost of this run.
Illustrated sample. The button steps through this example run; nothing runs here.
Architecture
Architecture maps how your product fits together, helps you find single points of failure before production does, estimates what it costs, and turns the plan into a real project.

ArchitectureIllustrated sampleChoose a component to take it down.
Example estimate
Launch setup: about $235 a month.
From typical AWS pricing for each component. Actual costs vary by region, usage and reserved instances.
Generated from the diagram
Plus Terraform, Kubernetes manifests and a Makefile.
cp .env.example .envdocker compose up -d --builddocker compose psReady for GitHub
Codelit can push this scaffold to a new repository in one commit.
Models
Start free. Choose what runs the work, and see where your prompt goes.

Run Qwen3 locally for short text tasks on compatible computers. Download once, then run without per-token AI fees.
Model size
Qwen3 Balanced: 1.7B parameters, about 1 GB download.
Experimental browser AI. Needs GPU support and runs while this tab is open.
Local inference stays on your device. Chats still sync to your Codelit account. Hosted tools and AI model costs have separate limits.
Integrations
Connect what you use. Keep the work in one place.
Also connects with
MCP
Draft product plans, architecture, and Agent Teams with public planning templates. The MCP server is live; the ChatGPT directory listing is submitted for review and is not yet available.
Real ChatGPT Developer Mode demo: list four workflows, fetch the Product Plan template, and draft a book-swap MVP with two acceptance tests. Nothing is saved or deployed. Silent recording with shortened pauses.

Codelit for Mac
Keep conversations, memory and skills together. Free, with no account required. Download a local model once, then chat offline.
Summarize my notes from today.
The launch moved to Friday. Design needs one more review, and Sam owns the release notes.
Local model on this Mac
Apple silicon, macOS 14 or later, 8 GB memory minimum. Intel Macs are not supported. The Mac App Store app does not include browser automation, computer control or background routines. Those web features remain in Codelit online.

Start with free local or cloud AI, or bring your own models.
Hosted browser and computer usage has separate limits.
Choose an app to see its window.

Matched onshippreview
Or start from a proven outcome
How a Team starts
Best match: Release Safety Team.

Turn a scoped issue into a branch, preview, browser check, pull request, and release proof.
Works with
You get back

Includes the estimated cost of this run.
Illustrated sample. The button steps through this example run; nothing runs here.
Choose a component to take it down.
Example estimate
Launch setup: about $235 a month.
From typical AWS pricing for each component. Actual costs vary by region, usage and reserved instances.
Generated from the diagram
Plus Terraform, Kubernetes manifests and a Makefile.
cp .env.example .envdocker compose up -d --builddocker compose psReady for GitHub
Codelit can push this scaffold to a new repository in one commit.

Run Qwen3 locally for short text tasks on compatible computers. Download once, then run without per-token AI fees.
Model size
Qwen3 Balanced: 1.7B parameters, about 1 GB download.
Experimental browser AI. Needs GPU support and runs while this tab is open.