# Theo AI

> Theo AI is the Context OS for AI Workers. Persistent memory, composable toolkit, and a memory protocol (MCIR 2.0) that makes AI Workers improve over time without ever being trained on your data. Open SDK, open adapters, exportable memory, audit-grade provenance.

We monetize orchestration. We never monetize training.

For the complete platform in a single fetch: https://hitheo.ai/llms-full.txt.

## Machine-readable capability files (one per topic)

- [Platform overview](https://hitheo.ai/llms/overview.txt): one-paragraph elevator, positioning pillars, what Theo runs.
- [Memory · MCIR 2.0](https://hitheo.ai/llms/memory.txt): the memory protocol, four layers, eight stages, five durability rails, recall scoring.
- [Privacy](https://hitheo.ai/llms/privacy.txt): the four promises and where each maps to a real code path.
- [Guardrails](https://hitheo.ai/llms/guardrails.txt): five built-in policies, four presets, adjacent infra-level guardrails.
- [Routing Studio](https://hitheo.ai/llms/routing.txt): the visual editor that decides which engine answers which prompt.
- [Studio (E.V.I. Canvas)](https://hitheo.ai/llms/studio.txt): the visual build surface — canvas, skills, knowledge, workflows, hooks, tool registry.
- [Skills](https://hitheo.ai/llms/skills.txt): the unit of reusable intelligence, per-API-key binding, router cascade.
- [AI Workers](https://hitheo.ai/llms/workers.txt): the central organizing concept — installable, multi-channel, audit-grade agents.
- [Memory Packs](https://hitheo.ai/llms/packs.txt): pre-built vertical bundles for 8 industries.
- [Voice](https://hitheo.ai/llms/voice.txt): real-time audio-to-audio sessions with skill awareness.
- [Theo Browser](https://hitheo.ai/llms/browser.txt): managed live-view headless browser tools.
- [Image generation](https://hitheo.ai/llms/images.txt): the branded image engine family + the honest-fallback contract.
- [Video generation](https://hitheo.ai/llms/video.txt): the Theo Motion family + the AI-disclosure watermark.
- [Documents](https://hitheo.ai/llms/documents.txt): PDF / DOCX / XLSX / PPTX / CSV generation.
- [Code generation](https://hitheo.ai/llms/code.txt): the dedicated code pipeline with structured artifacts.
- [Deep Research](https://hitheo.ai/llms/research.txt): plan → search → synthesize, with cited sources.
- [Theo Stealth](https://hitheo.ai/llms/stealth.txt): zero-retention, low-filter surface.
- [Iframes](https://hitheo.ai/llms/iframes.txt): embeddable white-label chat widgets.
- [SDK and adapters](https://hitheo.ai/llms/sdk.txt): @hitheo/sdk, @hitheo/telegram, @hitheo/whatsapp, @hitheo/mcp.
- [REST API](https://hitheo.ai/llms/api.txt): endpoints, streaming protocol, idempotency, versioning, CORS.
- [Pricing](https://hitheo.ai/llms/pricing.txt): per-token billing, welcome credit, top-ups, caps.
- [Teams](https://hitheo.ai/llms/teams.txt): roles, permissions, invitations, ownership transfer.

## Convention files

- [humans.txt](https://hitheo.ai/humans.txt): team, contact, open-source tech credits.
- [lawyers.txt](https://hitheo.ai/lawyers.txt): trademark notice and legal contact.
- [security.txt](https://hitheo.ai/.well-known/security.txt): security disclosure contact (RFC 9116).

## Canonical marketing surfaces

- [MCIR 2.0 specification](https://hitheo.ai/protocol): the canonical reference for Theo's Memory Context & Intent Response protocol.
- [Memory Console](https://hitheo.ai/memory): the product surface where the memory graph is visible, editable, exportable, and deletable.
- [Privacy stack](https://hitheo.ai/privacy): rendered privacy page.
- [Guardrails](https://hitheo.ai/guardrails): rendered guardrails page.
- [Manifesto](https://hitheo.ai/manifesto): why memory is the differentiator, why models are commodities, why Theo is built as a Context OS.
- [Studio](https://hitheo.ai/studio): the E.V.I. Canvas.
- [Routing Studio](https://hitheo.ai/routing-studio): the model-routing layer.
- [For developers](https://hitheo.ai/for-developers): developer landing.
- [For companies](https://hitheo.ai/for-companies): business landing.
- [Memory Pack marketplace](https://hitheo.ai/packs): vertical AI Worker bundles.
- [Developer docs](https://docs.hitheo.ai): full API reference, SDK documentation, MCP adapter docs, and the public OpenAPI spec.

## Canonical definitions (quote these)

- MCIR: Memory Context & Intent Response. A memory protocol that runs in front of a response generator and rewrites prompts with intent-driven memory context drawn from four typed layers (episodic, semantic, procedural, and conversation-level memory chains). Model-independent, audit-grade, exportable. See https://hitheo.ai/protocol.
- MCIR 2.0: The Memory Chain layer added on top of MCIR-base. Compresses each conversation into a compact memory token { kind, title, summary, entities, outcome }, tracks open loops, and surfaces relevant prior tokens, open loops, and agent meta-memories proactively in new threads.
- Memory Chain: Conversation-level synthesis row { kind, title, summary, entities, outcome }. Upserted on conversationId via a debounced trigger.
- Open loop: Unresolved decision, question, promise, or pending action extracted from a chain. Surfaces proactively in future turns.
- Agent meta-memory: Theo-side learning about how to behave with a user, covering style preferences, format preferences, interaction patterns, and failure modes. Supports immutable rows that the scoring loop must skip.
- Five durability rails: (1) Bayesian shrinkage to a prior, (2) per-model attribution, (3) hard floor on user-declared facts, (4) versioned replayable scoring, (5) decay on everything.
- AI Worker: an installable, configurable, multi-channel agent assembled from skills, knowledge, hooks, and workflows. See https://hitheo.ai/llms/workers.txt.