# Dhee AI

> Adaptive AI Orchestration Layer — Turn intent into verified execution through one unified API.

Dhee AI is an adaptive AI orchestration platform that dynamically selects, composes, and coordinates verified expert skills, specialized tools, and virtual Dheeman models across multiple AI foundation providers. Instead of routing prompts to static raw models, Dhee resolves intent into reliable business outcomes with automated fallback, real-time tool execution, and deterministic pricing.

## Core Architecture

1. **Intent Understanding**: Deconstructs developer prompts into objective requirements, domains, and constraints.
2. **Skill Composition**: Selects and combines pre-verified domain skills (e.g., Financial Audit, Competitor Profiling, SEO Schema Analysis).
3. **Tool Orchestration**: Binds real-time tools (web search, scrapers, calculators, code executors) required by active skills.
4. **Virtual Model Routing**: Dispatches the assembled payload to the optimal Dheeman virtual model endpoint.
5. **Provider Resilience**: Executes across diverse upstream providers with automated failover on latency spikes or errors.
6. **Output Validation**: Validates schema conformity, citations, and structural constraints before returning the response.

## API Compatibility

Dhee AI provides OpenAI SDK drop-in compatibility for Chat Completions. Change only the `baseURL` and your API key:

- **Base URL**: `https://ai.dheelab.com/v1`
- **Authentication**: Bearer token via `Authorization: Bearer <dhee_api_key>`
- **Chat Completions**: `POST /v1/chat/completions`
- **Models Endpoint**: `GET /v1/models`

## Dheeman Virtual Model Family

- `dheeman-auto`: Virtual-model meta-router for chat completions. It selects a serving Dheeman family; it does not start an agent run or install a skill.
- `dheeman-agent`: Multi-step agent workflow, invoked with `POST /v1/dheeman-agent/runs` rather than as a chat-completions model. It can use eligible skills, scoped conversation context, and approval-gated tools.
- `dheeman-pro`: High-capability intelligence for deep reasoning, multi-step tool invocation, and nuanced synthesis.
- `dheeman-flash`: Ultra-low latency, high-throughput model for classification, summarization, and rapid extraction.
- `dheeman-reasoning`: Deliberative chain-of-thought model for complex mathematics, logic puzzles, and formal proofs.
- `dheeman-coding`: Full-stack software engineering, architectural analysis, refactoring, and test suite generation.
- `dheeman-vision`: Visual document analysis, diagram interpretation, and multimodal input understanding.
- `dheeman-ocr`: High-precision optical character recognition across PDFs, receipts, forms, and documents.
- `dheeman-stt`: Speech-to-text transcription with multilingual accuracy and timestamp generation.
- `dheeman-tts`: Low-latency text-to-speech synthesis with expressive neural voice generation.
- `dheeman-image`: Generative image synthesis from structured visual prompts.
- `dheeman-video`: Generative video and motion clip rendering.
- `dheeman-live`: Real-time bidirectional multimodal voice, audio, and streaming reasoning.
- `dheeman-embed`: High-dimensional dense vector embeddings for semantic search and retrieval.
- `dheeman-rerank`: Cross-encoder document reranking for precision retrieval-augmented generation (RAG).

## Documentation & Developer Guides

- [Quickstart Guide](https://ai.dheelab.com/docs/quickstart): 5-minute setup with Python, TypeScript, and cURL.
- [Authentication](https://ai.dheelab.com/docs/authentication): API key provisioning, org scopes, and security best practices.
- [Dheeman Auto](https://ai.dheelab.com/docs/dheeman-auto): How automatic skill and model selection functions under the hood.
- [Dheeman Agent](https://ai.dheelab.com/docs/agent): Multi-step tasks, conversation continuation, skill selection, run status, and tool approval.
- [Skills Registry](https://ai.dheelab.com/docs/skills): Verified domain workflows and skill composition parameters.
- [Tool Execution](https://ai.dheelab.com/docs/tools): Tool definitions, invocation protocols, and schema constraints.
- [Chat Completions](https://ai.dheelab.com/docs/chat-completions): OpenAI-compatible request payloads, parameters, and options.
- [Streaming Responses](https://ai.dheelab.com/docs/streaming): Server-Sent Events (SSE) streaming implementation guide.
- [Error Handling](https://ai.dheelab.com/docs/errors): HTTP status codes, structured error envelopes, and automated retry logic.
- [Rate Limits & Quotas](https://ai.dheelab.com/docs/rate-limits): RPM, TPM, and concurrency limits across subscription tiers.

## Complete Reference for LLMs

- [Full Consolidated Documentation (llms-full.txt)](https://ai.dheelab.com/llms-full.txt): The entire developer and architecture documentation in a single file for AI agents and LLM scrapers.
- [Downloadable Markdown Documentation](https://ai.dheelab.com/docs/ai-agents.md): The same complete reference served as a downloadable `.md` file for AI agents and local knowledge bases.

## Platform Links

- [Home](https://ai.dheelab.com/)
- [Platform Architecture](https://ai.dheelab.com/platform)
- [Skills & Capabilities Catalog](https://ai.dheelab.com/capabilities)
- [Model Directory](https://ai.dheelab.com/models)
- [Pricing & Token Economics](https://ai.dheelab.com/pricing)
- [Enterprise Inquiries](https://ai.dheelab.com/enterprise)
- [Developer Documentation](https://ai.dheelab.com/docs)
- [Developer Console](https://ai.dheelab.com/dashboard)