Skip to content
The serverless runtime for AI agents

Build agents your way.
Theo runs the rest

You design how the agent thinks. The memory, knowledge, guardrails, observability, and scale around it are a whole AI platform — Theo runs it all. Live in ~7s.

Open-source primitives · Managed or self-hosted · Any LLM

Deploy to TheoCloud
Or start in your terminalTheoKit · TypeScript framework for AI agents · Apache-2.0
$ npx create-theokit my-app
the agent stack → one runtime
LangChain
Pinecone
m0Mem0
LSLangSmith
OpenAI
Vercel
CrewAI
WvWeaviate
ZepZep
LfLangfuse
GRGuardrails
Anthropic
Modal
Temporal
memory
knowledge
guardrails
observability
execution
scale
Theo
The wrong foundation

The agent was the goal.The infrastructure became the work.

Every layer of today's cloud was designed for web apps — request in, response out, stateless, a human on the other end. An agent is none of those things. It remembers, retrieves, decides, and acts.

Read the manifesto

Any model · no lock-in

Bring any model. Switch in one line.

OpenAI, Anthropic, Google, Mistral, Llama, Ollama — 28 providers, swap anytime without rewriting your agent.

OpenAIAnthropicGeminiLlamaMistralOllamaHugging FacePerplexityNVIDIADatabricksReplicatexAI

Apache-2.0 · TypeScript-first · open-core

The complete runtime

Everything built in. One runtime.

01

Agents

Define an agent, give it tools, and ship the whole loop in a few lines of TypeScript.

Apache-2.0TypeScript
agent.ts
1import { Agent, defineTool } from "@theokit/sdk";
2import { z } from "zod";
 
4const refund = defineTool({
5 name: "refund_order",
6 description: "Refund an order by id",
7 inputSchema: z.object({ orderId: z.string() }),
8 handler: async ({ orderId }) => {
9 await billing.refund(orderId);
10 return `Refunded ${orderId}`;
11 },
12});
 
14const agent = await Agent.create({
15 model: "claude-sonnet-4-6",
16 tools: [refund],
17});
 
19const run = await agent.send("Refund order #8814");
live
agents

one agent · every capability in orbit

Everything in Agents8 features
Agent in 6 lines

The whole agent loop, in 6 lines.

No orchestration framework to wire up — Agent.create gives it tools, agent.send runs it.

Apache-2.0TypeScript

Start building

Star on GitHubRead the docs

Apache-2.0. Your code, your repository.

What your user sees

Your agent, handling a real request.

One conversation. Your user gets clean answers — you get memory, retrieval, real actions and guardrails, in one runtime.

What your user seeshelp.acme.com
Message support…
What you shiptrace · agent.run

Simple for them. The whole stack for you — one runtime, not five vendors.

How it works

Three verbs. Build. Ship. Scale.

From npx create-theokit to a live URL to production — one workflow.

  1. 01

    Build

    Scaffold a working agent — memory, search and chat UI, built in.

    $npx create-theokit my-app
    ✓ agent + memory + search wired
    $npm run dev → localhost:5173
    Apache-2.0local devTypeScript
  2. 02

    Ship

    Push to TheoCloud. Live URL in minutes, rollback in seconds.

    $theo deploy
    building · verifying · going live…
    ✓ Live at my-app.theo.dev
    live in minutesrollback in secondsreal Postgres + queues
  3. 03

    Scale

    Switch models, watch every call, keep state — without a redeploy.

    LLM calls12.4k today
    modelclaude-sonnet-4-6 ⇄
    sessions23 · state kept
    BYOKevery LLM callyour data

Commercial · Early access

TheoDB. The AI database.

Built on Postgres — embeddings, vector search, generation, ranking and natural-language queries as SQL functions. One database for your data and your AI — no ETL to a separate vector store. The data layer under TheoCloud, also sold on its own.

theodb · psql
connected
-- semantic search — ranked by meaning, in SQL
theodb=#

Everything, one database

The full AI surface — embeddings to natural-language queries — in the database your ORM already speaks.

Indexes

HNSWIVFFlatDiskANNSBQ / PQ quantization

Search

semantichybrid · FTS + vector (RRF)filtered

AI in SQL

embedgeneratebatchranksentimentsummarizeNL → SQL

Postgres-native

100% Postgres-compatibleany modelany driver & ORMno ETL

One database for your data, your vectors, and your AI — no glue, no second system.

Pricing

Pricing scales with usage. Start free, pay as you grow.

Start free, upgrade when an agent is doing real work. Every tier uses your own AI provider keys — Theo never sees your prompt.

Hobby

Single agent in production

Free
  • 1 agent · 100 builds/mo
  • *.theo.app subdomain · 1 team member
  • 1 month of Starter included on signup
Start free
Recommended to start

Starter

Early production agent

$19/mo
  • 3 agents · 500 builds/mo
  • Custom domains · unlimited team
  • 30-day observability retention
Start free

Growth

Sustained user traffic

$79/mo
  • 10 agents · 2,000 builds/mo
  • 60-day observability retention
  • Unlimited team
Start free

Pro

Teams · multiple agents

$249/mo
  • 25 agents · 10,000 builds/mo
  • 90-day observability retention
  • Priority email support
Start free
No card to start·BYOK — Theo never sees your prompt·Cancel anytime

An agent is one deployed agent app — its own URL, sessions, and memory. 1 build = one deploy: each `theo deploy` you run, or each push that triggers an auto-deploy. Memory and search limits are per month — full quotas in the plan matrix.

See the full plan matrix →

The runtime for AI Native Companies

You build the agent. Theo runs the rest.

Hobby is free forever. Your code, your keys, your data.

No card required. Cancel anytime on paid tiers.

Or for developers
Start in your terminalTheoKit · TypeScript framework for AI agents · Apache-2.0
$ npx create-theokit my-app