The backend you'd build anyway.

Memory. LLM routing. Permissions. Cost tracking.Production agent infrastructure, ready now.

Documentation

What you get

Infrastructure, not abstractions

LLM Proxy

Route to any model. One endpoint, every provider.

Hosted Memory

Three-tier memory. User, agent, session. No database setup.

Sandboxing

iOS-style permissions. Control what agents can access.

Cost Tracking

Per-request logging. Know exactly what each run costs.

How you integrate

One agent, anywhere

API
await halo.run("researcher", {
  prompt: "Find competitors"
})
React
const { run } = useAgent("researcher")

run("Find competitors")
CLI
$ halo run researcher \
  "Find competitors"

Why agents

Not another workflow tool

Workflows
  • If X happens, do Y
  • You design every step
  • Breaks on edge cases
Agents
  • Given a goal, figure out how
  • Agent decides the path
  • Adapts and reasons

Get early access

API access is opening soon. Join the waitlist.

Already building? Publish an agent