Approval workflows
Pause execution until a human approves. Configure per-tool: owner only, any member, or self-approval. Modify parameters before proceeding.
Stop building internal tools. Just talk to your AI. It handles the work, generates the UI, and asks when it needs you.
Connects to
Problem
Nobody wants to maintain someone else's Retool app. The original author left. The API got deprecated. And now you're stuck debugging JavaScript embedded in a drag-and-drop canvas.
Revenue
$42,350$--,---
Users
1,247NaN
Orders
892undefined
Growth
+12%
Monthly Revenue
Failed to loadRecent Users
| Order | Customer | Amount | Status |
|---|---|---|---|
| #4521 | Acme Corp | $2,400 | Paid |
| #4520#---- | TechStart Inc[Object] | $1,200$NaN | PendingError |
| #4519 | StartupXYZ | $99 | Paid |
What if you never had to build—or maintain—internal tools again?
The solution
Just tell your AI what you need. It generates the UI, fetches the data, and asks when it needs you.
The agent looks up the order, checks the customer history, and generates a refund form with all the details pre-filled. You just review and approve.
Process a refund for order #4521. Customer says item arrived damaged.
Found the order. Here's the refund form:
The agent queries your database, displays the results in a sortable table, and lets you select rows for bulk actions. No SQL knowledge required.
Show me all failed payments from this week
Found 5 failed payments. Select rows to retry:
2 of 5 selected
The agent pulls data from Stripe, visualizes it with charts, and asks if you want to export or share with the team. Full report, zero manual work.
Generate the monthly revenue report for December
Here's your December revenue report:
Zero maintenance: Every interface is generated fresh. No dashboards to update. No forms to maintain. The AI creates exactly what you need, when you need it.
But there's more
Like a colleague tapping your shoulder. Set triggers—cron, webhooks, events—and your agents reach out when something needs attention. You respond when you're ready.
3 tickets overnight — review drafts?
Acme Corp payment failed — send grace email?
PR #421 merged — tests passed
3 tickets overnight — review drafts?
Acme Corp payment failed — send grace email?
PR #421 merged — tests passed
Your turn
Chat with Copilot, it writes the agent. Pure JavaScript you can read, edit, and extend.
Just ask
Talk to Copilot like a colleague. It understands your intent, proposes changes, and generates tools in real-time. Review the diff, then apply or reject—you're always in control.
Give it superpowers
Every tool is a JavaScript function with typed params and returns. Use context.resources to query Postgres, call Stripe, and more.
// Params: { hours: number }// Returns: { payments: Payment[] } const { hours } = paramsconst stripe = context.resources.stripe const payments = await stripe.paymentIntents.list({ created: { gte: Date.now() - hours * 3600000 }, limit: 100}) const failed = payments.data.filter( p => p.status === "requires_payment_method") return { payments: failed } Ship it
Choose your LLM (OpenAI, Anthropic, Google). Set which tools require human approval. Configure triggers—cron, webhooks, or events. One-click deploy with semantic versioning.
Zero maintenance
Forms, tables, charts, approvals—generated fresh for each task. Nothing to maintain.
Forms
Text, numbers, dates, dropdowns, checkboxes—whatever the task needs, with built-in validation.
Tables
Sortable, selectable rows. Pick one or many for bulk operations. No configuration needed.
Charts
Line, bar, pie—visualize any data instantly. No BI tool required.
Approvals
Confirm, modify, or reject. Dangerous operations require human sign-off.
Result: Your team talks. AI handles the UI. No Retool. No React dashboards.
Production ready
Pause execution until a human approves. Configure per-tool: owner only, any member, or self-approval. Modify parameters before proceeding.
Role-based access control at the tool level. Define who can run what, and which operations require approval.
Connect Postgres and MySQL from your VPC. Deploy a lightweight connector, outbound-only. No firewall changes, no exposed ports.
Workflows survive restarts, deploys, and outages. State recovery is automatic. Long-running tasks handled gracefully.
OpenAI, Anthropic, Google. Use the right model for each agent. Switch providers without changing your tools.
Full Vim keybindings in the code editor. Because real developers don't use mice. Escape to normal mode, you know the rest.
Full transparency
Self-host or use our cloud. Audit every line. No vendor lock-in.
╭─ synatra ────────────────────────────────╮ │ │ │ License Apache 2.0 │ │ Stack SolidJS · Temporal · Hono │ │ Stars ★ --- │ │ │ ╰──────────────────────────────────────────╯
∧_∧
(^ω^)
| |>
U U
Ready?
Deploy your first AI agent in minutes. Free to start, open source forever.
Open source. Self-host or use our cloud.