MCIR 2.0 memory, Guardrails, Routing Studio, Skills, Workflows, and Hooks, exposed as a clean SDK. Persistent memory, governed context, agentic tools, full audit. No vendor lock-in, no training on your data.
npm install @hitheo/sdk - then pick your use case.
OPEN SOURCE · TYPESCRIPT · FULL DOCS
DOCS: SDK REFERENCE →import { Theo } from "@hitheo/sdk";
const theo = new Theo({ apiKey: "theo_sk_..." });
const res = await theo.complete({
prompt: "Explain quantum computing simply",
mode: "auto", // → routes to Theo Flash (quick chat)
});
console.log(res.text);
console.log(res.model); // "theo-1-flash"
console.log(res.cost); // "$0.0003"Same brain, same memory, same skills. Every channel. Open-source adapters.
THE MCIR 2.0 PROTOCOL
Eight stages decide what Theo remembers, what it ignores, and how it proves both. The same protocol governs every AI Worker on the platform.
CLICK ANY STAGE TO LEARN MORE
READ THE MCIR 2.0 SPEC →The right model for every message. Every time. Automatically.
Click any model to explore
Docs: Engine Routing →Theo reads every prompt and picks the right engine in the background. No keywords, no configuration, no routing tables to maintain.
Simple questions use fast, cheap models. Complex reasoning gets the premium ones. You save on every routine request, automatically.
Most prompts classify in milliseconds. Quick answers feel instant. Long-running tasks stream so users never wonder if it froze.
If one provider hiccups, Theo slides to the next. Your widget, your app, and your automations keep working. Your users never notice.
Theo's own cross-model analysis across the public benchmarks developers use to evaluate frontier models: coding, reasoning, multimodal, and agentic work.
| Benchmark | Claude Opus 4.6Anthropic | Claude Sonnet 4.6Anthropic | GPT-5.3 CodexOpenAI | Gemini 3.1 ProGoogle | Grok 4.1xAI | DeepSeek V3.2DeepSeek | theoOrchestrated |
|---|---|---|---|---|---|---|---|
SWE-bench Verified Resolves real GitHub issues end-to-end against a hidden test suite. | 80.8% | 79.6% | 80% | 80.6% | 75% | 67.8% | 80.8% |
Aider Polyglot Edits code across six languages with self-correction on failed tests. | 85% | 82% | 80% | 78% | 79.6% | 74.2% | 85% |
Reasoning leaders trail on agentic work. Coding leaders trail on math. Multimodal leaders trail on tool use. Pin yourself to one vendor and you inherit every weakness on the row that didn't go their way. Theo evaluates across the table and routes each request to whichever model wins on this kind of task. The theo column is the ceiling that orchestration reaches.
These numbers reflect Theo's own analysis using each linked benchmark as an evaluation framework. They aren't official scores certified by the benchmark organizations or model vendors. The Theo column is derived as the row maximum across the listed competitors.
Type a prompt and watch theo decompose it into a multi-model execution chain, routed and run in real time.
TRY A MULTI-MODEL PRESET OR TYPE YOUR OWN
// The same routing happens whether you call it from your app,
// embed the widget, or type in the playground.
theo.ask({
prompt: "Build me a video about a dog on a skateboard and roast me about it",
mode: "auto",
});
// → Theo Code (auto)E.V.I. is the canvas. An AI Worker is what you ship with it, with your persona, your skills, your knowledge, and theo's engine under the hood.
TOUR THE BUILD LAYER →Replace Theo's personality with your own brand. Your customers see your AI Worker, never Theo himself.
Install marketplace skills for domain expertise. Pass inline tools for your product's actions.
Drop PDFs, DOCX, XLSX, CSV, MD onto a skill via the Knowledge Node. 10MB/file · 100/skill · priority-weighted. Never trained on.
Drag-and-drop builder. Connect prompts, models, tools, conditionals, vision nodes. Compile, test, version, publish.
MCIR memory, routing, guardrails, intent classification, agent loop, billing, caching, audit , all handled.
const model = pickModel(prompt);
const tools = resolveTools(user);
const system = buildPrompt(persona, skills);
const res = await llmProvider.complete({ model, ... });
const toolResults = await executeTools(res);
const final = await llmProvider.complete({ ... });
await debitCredits(user, res.usage);
await logAudit(user, res);import { Theo } from "@hitheo/sdk";
const theo = new Theo({ apiKey: process.env.THEO_KEY });
const evi = theo.evi({
persona: "You are Eva, AcmeCorp's AI...",
skills: ["insurance-quoting"],
tools: [lookupCustomer, updatePolicy],
});
const res = await evi.complete({ prompt: "..." });ONE API CALL. EVERYTHING HANDLED.
Intent classification. Model routing. Skill injection. Tool execution. Memory. Streaming. Billing. Audit. Caching. All from evi.complete().
VERTICAL MEMORY PACKS
Pre-built vertical skills with the entities, behaviors, and escalation rules your industry already expects. Install in a click , fork to customize, publish your own.
INSTALL · FORK · PUBLISH
Applicants, dependents, carriers, plans, enrollment periods. Licensed-agent escalation built in. Never binds coverage.
applicant · broker · policy · carrier · plan
Clients, treatments, consultations, contraindications. Books consults, never makes medical claims, escalates clinical questions.
client · treatment · consultation · provider
Accounts, users, tickets, plan tiers, integrations. Severity-aware escalation, ticket linking, and integration-specific runbooks.
account · user · ticket · plan tier · integration
Workspaces, roles, onboarding stages, activation events. Detects blockers, celebrates milestones, never spams “stuck” users.
workspace · role · stage · activation · blocker
Clients, matters, jurisdictions, deadlines. Preserves chronology, logs disclaimers, escalates attorney-required questions.
client · matter · jurisdiction · deadline
Plus Real Estate, Healthcare Front Desk, Recruiting / Staffing.
import { defineSkill } from "@hitheo/sdk";
export default defineSkill({
name: "property-management",
entities: ["tenant", "unit", "lease"],
tools: [openTicket, scheduleVisit],
knowledge: ["./policies/*.pdf"],
});
// npx theo publishDrag-and-drop builder. Connect prompts, models, tools, knowledge, and vision nodes. No code required.
OPEN CANVAS →PUBLISH YOUR WAY
Publish privately (just you), to your organization (team-only), or to the public marketplace. Earn credits for every API call that activates your pack.
One dashboard for everything. Keys, usage, logs, routing config. No more switching between 4 provider consoles.
Channel-agnostic core. Open-source adapters for Telegram and WhatsApp forward webhook updates into Theo; pass the same conversation_id to share memory across channels.
@hitheo/telegram@hitheo/whatsapp@hitheo/sdk@hitheo/telegram@hitheo/whatsapp@hitheo/sdkslack-messenger skill or incoming webhooktheo.completeEvery action in Theo fires a signed event. Route it to Slack, email, your CRM, or any of 5,000+ apps, without writing a backend.
Stream every completion with mode, tokens, and cost into your warehouse.
Theo-Webhook-Signature: v1=…{
"event": "completion.created",
"mode": "fast",
"tokens": 340,
"costCents": 2
}Already shipping with the OpenAI SDK? Point baseURL atapi.hitheo.aiand you're on Theo. Auto-routing, multi-model, same API.
import OpenAI from "openai";
const theo = new OpenAI({
apiKey: process.env.THEO_API_KEY,
baseURL: "https://api.hitheo.ai/v1",
});
const res = await theo.chat.completions.create({
model: "theo-1-auto",
messages: [{ role: "user", content: "Write me a haiku about Miami" }],
stream: true,
});Not a toy. Not a wrapper. Theo is the same infra our own apps run on.
Token-level streaming with tool events, artifacts, and follow-ups. Also speaks the OpenAI chat.completion format natively.
Skills declare their tools in a manifest. Theo plans, calls, observes, and retries with full audit trails.
HMAC-signed deliveries with automatic retries, dead-letter queue, and idempotency keys for safe replays.
Circuit-breaker across OpenRouter, Anthropic, Gemini, and Venice. Users never notice an outage.
Pass a conversation_id and every channel (web, Telegram, WhatsApp) shares the same memory.
Per-token passthrough with unified billing. Daily caps, per-key attribution, and soft/hard limits.
Pricing
Add funds. Use the API. No plans, no subscriptions, no commitments.
Passthrough model costs plus a small platform fee.
Add dollars to your balance and draw down per token. Top up any time.
Create accountBilling by OpenCharts Inc., the company behind Theo.
One API key. Persistent memory. Composable skills. Guardrails out of the box.
Start free in 30 seconds. No credit card required.