Ship with agents. Keep the final say.
Describe the outcome once. Product Plans, Architectures, Agent Teams, supervised runs, and receipts stay in one Thread, where you can inspect or edit any result without losing the conversation.
Free · no signup · no API key · no model spend
$0
Sample model spend
Read-only
First real proof
Exact
Write preview
Recoverable
Partial runs
Durable
Run Receipt
Release Safety proof
See exactly what happened.
From pinned issue to verified preview, every stop, decision, and receipt stays visible.
Proof 01
Ground the change
Pin one GitHub issue, prepare the reviewed preview, and verify that exact URL in the browser.
GitHub issueScope pinned
Browser QAPreview verified
release.proof -> preview verified
DOM and screenshot tied to the exact deployment
Proof 02
Stop without losing work
Keep completed evidence, halt before a repeated action, and preview the smallest repair.
Failed checkRun stopped
Repair previewYour call
run.stopped -> repair preview
Two receipts preserved with zero repeated writes
Proof 03
Carry proof forward
Save the exact decision in a Run Receipt, then accept it into a delivery-ready plan.
Run ReceiptProof linked
Plan & ShipWork ready
receipt.accepted -> delivery plan created
Sources, decisions, checks, and next owners retained
Run the Release Safety Sample
Committed sample data · no signup · no API key · no external writes
Flow · Activity
Simple first. Visible when needed.
Use Ask Codelit for ordinary work. Flow keeps responsibilities, tools, handoffs, and approvals directly editable; Activity keeps every run, decision, and receipt.
Browser · Local
Use the web. Keep local work local.
Give an Agent a managed browser when a real site is the only interface, or run with reviewed files and your key in the active tab. Both paths expose exactly where work runs and when a human must take control.
Browser findings that reach your team
The Agent verifies a site with current DOM and screenshot evidence. After approval, the next Agent can post that verified result through Slack or another connected app and retain both receipts.
Committed sample data · zero external calls
Local runs, no Codelit compute
Selected files, isolated browser orchestration, and local checkpoints stay on the device. The user's provider key pays only for model calls.
Context in. Runnable team out.
Ground every decision in the work that already exists, then turn it into tools, boundaries, approvals, and an engineering-ready handoff.
Repository
Code, policies, runbooks
Product context
Figma, tickets, docs
Runtime
Terraform, K8s, Compose
Break it here. Not in production.
Inject failures, trace the blast radius, and verify that human gates stop consequential writes before real tools or customers are involved.
Safe test
Dry run makes no model calls or external writes.
Seven audit paths
Security, compliance, performance, cost, CVEs, code quality, and load.
Evidence retained
Keep the failure, decision, approval, retry plan, and outcome in one trace.
One review. Every team.
Share the projected run before implementation, then carry the approved policy, tests, evals, and reference trace into engineering.
Agent launch review
Refund resolution agent
9/10
Review
Run path + gates
Protected
Share
Secure evidence link
Ready
Handoff
Tests + evals + trace
6 files
Any model. Same control plane.
Bring the provider you trust, choose models per agent, and keep team policy and run evidence independent from the model vendor.
OpenRouter
OpenAI
Anthropic
Gemini
xAI
Mistral
Groq
Perplexity
Together
DeepSeek
Cerebras
Connected where the work happens
Apps
GitHub
GitLab
Bitbucket
Figma
Jira
Notion
Linear
Slack
Vercel
Built-in capabilities
Google Workspace
Microsoft 365
Zendesk
Intercom
Stripe
Shopify
HubSpot
Salesforce
Sentry
PostHog
Datadog
PagerDuty
Supabase
Hosted automations
Start from schedules, app events, or signed webhooks.
Inbound webhooks
Trigger from Zapier, Make, n8n, or your own app.
Chat approvals
Route decisions back into the recorded trace.
Bring your own key
Run in the active tab with your provider key and zero Codelit model spend.
Prove one release before you connect production.
Run the complete Release Safety Sample, inspect its approval boundary, and keep the receipt without signup or model spend.