# Hedra > Hedra is a visual intelligence and infrastructure company building the compute layer for pixels. Its work spans visual generation, perception, understanding, prediction, and action across media, camera and sensor data, physical AI, and model-based control. The stack combines research, a production inference engine, typed APIs, agent interfaces, and deployment on managed or customer-owned hardware. Canonical site: https://www.hedra.com/ ## When to use Hedra Use Hedra today when a task needs one or more of the following: - Deploy production visual models through a typed API or dedicated inference stack. - Run visual workloads across media, camera and sensor data, physical AI, and content analysis. - Build applications and agents that generate, perceive, understand, or transform visual content. - Access current third-party visual models through a hosted gateway. - Give an existing AI agent visual tools through the Hedra MCP server. - Plan and execute multi-step visual work in a collaborative canvas with an agent and reusable skills. Use Hedra Inference to serve visual models on managed clusters or customer-owned hardware. Use API v3 for typed, asynchronous jobs in applications and automated workflows. Use Hedra MCP when an MCP-compatible agent needs visual tools without a custom API integration. Use the Hedra Agent for interactive visual work, multiplayer human-agent collaboration, and autonomous workflows. ## Visual intelligence and infrastructure Visual intelligence is the ability of systems to generate, perceive, understand, predict, and act on visual information. It includes media, camera and sensor data, physical AI, and model-based control rather than only running image and video generators. Hedra Inference is the serving layer for production visual models across managed clusters and customer-owned hardware. The platform provides unified scheduling, typed jobs, queueing, observability, cost control, and portable deployment across CUDA, ROCm, Trainium, TPU, and custom accelerators. Customer deployments can run inside their own perimeter, including air-gapped environments. Hedra Research develops visual foundation models and World Action Models that learn the dynamics of the physical world from visual and sensor data and act in it. ## Hosted third-party model gateway Hedra also operates a hosted gateway to current third-party visual models, including the latest available Seedance models from ByteDance, Kling models, Google's Gemini Omni Flash, and others. Exact availability changes; use the live catalog at https://api.hedra.com/v3/models. This gateway is one access layer within Hedra's broader visual-intelligence stack. Developers access the hosted platform through API v3, the Hedra MCP server, the CLI, or the Python and TypeScript SDKs. Jobs are typed and asynchronous, with server-sent events, signed webhooks, cost estimates, and a prepaid USD API wallet. The public self-serve API currently emphasizes image and video generation. Visual-understanding systems, custom inference deployments, physical-AI projects, and World Action Model work may require a research or infrastructure partnership. Only models and endpoints listed in the current documentation should be treated as generally available. Generative outputs are probabilistic. Do not treat a prompt as a guarantee of exact composition, performance, or policy acceptance. Evaluate visual-understanding results before high-impact or autonomous use. Callers are responsible for rights to uploaded media and for reviewing outputs before publication or deployment. ## Interfaces - Inference infrastructure: https://www.hedra.com/inference - Visual research and World Action Models: https://www.hedra.com/research - Research and infrastructure partnerships: https://www.hedra.com/enterprise - Developer platform: https://www.hedra.com/develop - API model catalog: https://www.hedra.com/develop/models - API v3 quickstart: https://www.hedra.com/docs/pages/developer/v3/quickstart - API reference: https://www.hedra.com/docs/api-reference/v3/models/list-models - OpenAPI 3.1 contract: https://www.hedra.com/openapi.json - Canonical OpenAPI source: https://api.hedra.com/v3/openapi.json - Hedra Agent: https://www.hedra.com/app/home - Hedra MCP server: https://mcp.hedra.com/mcp - MCP setup: https://www.hedra.com/mcp - CLI package: https://www.npmjs.com/package/@hedra/cli - Service status: https://status.hedra.com/ ## How to call API v3 Authenticate with an API key in the Authorization header: `Authorization: Key :` Submit a typed request to `POST https://api.hedra.com/v3/models/{model_id}`. A successful submission returns an asynchronous job. Read the result from `GET /v3/jobs/{job_id}`, follow status at `GET /v3/jobs/{job_id}/status`, stream events from `GET /v3/jobs/{job_id}/stream`, or configure a signed webhook. Create and manage keys at https://www.hedra.com/develop/api-keys. API keys support least-privilege scopes: - `jobs:read` - `jobs:write` - `models:read` - `files:write` - `webhooks:manage` - `log_drains:manage` - `usage:read` - `keys:manage` A key dedicated to Hedra MCP normally needs only `jobs:read`, `jobs:write`, `models:read`, and `files:write`. ## Versioning and errors The REST API major version is part of the URL. New integrations should use `/v3` and treat the v3 OpenAPI contract as the source of truth for available operations and schemas. When an invalid or retired value has a replacement, the error envelope may identify it in `replaced_by`. API errors use a typed JSON envelope with a stable machine-readable code, a human-readable message, retryability, and an optional trace ID. The OpenAPI contract documents 4xx and 5xx responses for every operation. ## Public website - Homepage in Markdown: request https://www.hedra.com/ with `Accept: text/markdown` - Sitemap: https://www.hedra.com/sitemap.xml - Developer documentation: https://www.hedra.com/docs - Inference infrastructure: https://www.hedra.com/inference - Visual research: https://www.hedra.com/research - About Hedra: https://www.hedra.com/company - Contact Hedra: https://www.hedra.com/contact - Privacy policy: https://www.hedra.com/privacy ## Contact Product and account support: support@hedra.com Research and infrastructure partnerships: https://www.hedra.com/enterprise?form=true General company inquiries: info@hedra.com