Closed Beta

Browser automation
that actually works.

Context discovery framework.

Describe what you want in plain English. Colt discovers your app, executes the task on a live browser, and hands you a production Playwright script.

Get Early Access See Use Cases
colt
What You Can Automate
Real workflows. Not toy demos.
COLT automates the browser tasks your team does manually every week.
01

E2E Test Generation

Describe test scenarios in English. Colt executes them and exports pytest-compatible test files. Plug directly into CI/CD — no Selenium, no hand-written selectors.

pytestCI/CDregressioncoverage
# Generate a full test suite colt execute "Create user, verify in table, delete" \ --react --record test_user_crud.py # Run it in CI pytest test_user_crud.py --headed
02

Internal Tool Automation

Automate repetitive workflows on admin panels, CRMs, and dashboards. Colt navigates Salesforce, Zendesk, Retool — including Shadow DOM and iframe-heavy UIs that break other tools.

Salesforceadmin panelsShadow DOMiframes
# Automate a Salesforce workflow colt execute "Create a new lead with company Acme \ and status Qualified" --react --record lead.py # Replay it daily via cron python lead.py
03

Onboarding Flow Validation

Verify that signup, onboarding, and checkout flows work end-to-end after every deploy. Colt fills forms, handles multi-step wizards, and catches regressions before users do.

signup flowscheckoutmulti-stepdeploy checks
# Validate onboarding after deploy colt execute "Sign up with email test@co.com, \ complete profile setup, verify dashboard loads" \ --react --record test_onboarding.py
04

Data Entry and Bulk Operations

Generate typed Python functions for every form in your app. Import them, call them from scripts, loop over CSVs. Turn manual data entry into a single command.

codegentyped functionsbulk opsCSV import
# Generate functions from all discovered forms colt generate automation.py # Use in your own scripts from automation import create_user, update_billing await create_user(email="a@co.com", role="admin")
How It Works
Four phases. Persistent intelligence.
Colt learns your app once — then executes unlimited tasks from that knowledge.
01

Discover

Autonomous BFS crawl of your entire app.

  • Shadow DOM piercing
  • Cross-origin iframes
  • Infinite scroll detection
  • Modal + tab probing
  • Pagination guards
02

Index

Builds a persistent semantic knowledge base.

  • LLM state summaries
  • Vector embeddings
  • State transition graph
  • API schema capture
  • Form field mapping
03

Execute

ReAct agent drives the browser in real time.

  • Observe / Reason / Act
  • Self-healing fallbacks
  • Loop detection + abort
  • Token budget control
  • Dialog auto-handling
04

Record

Exports reusable, LLM-free code.

  • Playwright async scripts
  • Pytest test files
  • Typed Python functions
  • JSON execution traces
  • Step-by-step screenshots
Compare
What makes Colt different
Manual PlaywrightAI Browser AgentsColt
Selector authoringWrite by handNot neededNot needed
App knowledgeNoneStarts fresh each runPersistent knowledge base
Code outputYes — you wrote itNo outputScripts, tests, functions
Shadow DOM + iframesManual configLimitedFull piercing support
Self-healingFails on first errorVariesFallback chains + loop recovery
Runs without LLMYesNoExported code is standalone
Documentation
Everything you need to get started
Quickstart
CLI Reference
LLM Providers

Install

Python 3.10+ and Node.js required.

# Install Colt pip install colt-automation playwright install chromium # Set your LLM key export GROQ_API_KEY=your-key-here

Initialize

Generates config, output dirs, installs browsers.

colt init my-app \ --url http://localhost:3000 \ --llm groq

Discover

Crawls your app and builds the knowledge base.

# Headless colt discover # Watch in browser colt discover --no-headless

Execute + Record

Run a task and export a Playwright script.

colt execute \ "Create admin user" \ --react --record replay.py
colt init <name> --url <url> --llm <provider>Initialize project with config and browser setup
colt discover [-c config.yaml] [--no-headless]Crawl the app and build knowledge base
colt execute "task" --react [--record file.py]Execute a task with the ReAct agent
colt generate output.py [--dry-run]Generate typed functions from discovered forms
colt record trace.json output.py --format <fmt>Convert execution log to script / pytest / json / all
colt search "query" [-n 5]Search knowledge base for matching app states
ProviderDefault ModelEnv Variable
Groqllama-3.1-70b-versatileGROQ_API_KEY
OpenAIgpt-4oOPENAI_API_KEY
Anthropicclaude-sonnet-4-20250514ANTHROPIC_API_KEY
Ollamallama3.1 (local)--

Get early access

We're onboarding teams for closed beta.