# Facet > Facet builds a custom AI model for a task your product repeats. You describe the task, and Facet builds the model, tests it on your examples, runs it and keeps improving it. The model is yours. Facet is the autonomous model factory. Every version is checked before it goes live, and Facet keeps the model right as you correct it. Facet is made by Halfspace. You ask for a model where you already work: from Claude Code, Cursor, Codex or any MCP client (https://facet.halfspacelabs.com/mcp), or by mentioning @Facet in Slack. The web app is where you watch, review and approve. The agent works alongside the people who know the task: it writes the prompt and the evals, makes the training data, trains a small open-weight model, checks the result on hidden examples and puts live only what passes. It is meant to feel like your own frontier lab. No frontier lab will retrain its model because you asked. Facet will. Facet is open to invited teams first. To get access, join the waitlist at https://halfspacelabs.com/waitlist with the task you have in mind. ## Facet - What it does: you describe a task in a few sentences with a few examples, and Facet's agent builds a model for it. It drafts the rest and trains the first version for you. Every version is tested against your standard and rolls back in one step. - Quality: you define what good looks like as evals, plain statements of a good answer. Facet tests every new version against them on hidden examples, and after the first version an update goes live only if nothing the live version passes gets worse. - Feedback: flag a wrong answer, and Facet turns it into an eval and proposes a fix with its cost before anything trains. - Compare: score any version next to the frontier model you use today, on the same hidden examples, with the cost of 1,000 correct answers on each side. - Stays current: an owner can turn on a scheduled check that scores the live version on new examples and drafts a quoted retrain when the score drops. A person approves each one. - Use it anywhere: call it with any OpenAI-compatible SDK at https://api.halfspacelabs.com/v1, or export it and run it yourself. - Example tasks: support ticket routing, invoice and receipt extraction, lead scoring, personal data redaction, intent detection, log triage, content moderation, review tagging, resume parsing, orders from email. ## Use cases, comparisons and questions Each page is also served as Markdown at the same path with .md added. - [Support ticket routing](https://halfspacelabs.com/use-cases/support-ticket-routing.md): route every ticket to the right team with the right urgency - [Data extraction](https://halfspacelabs.com/use-cases/data-extraction.md): turn emails, invoices and contracts into clean fields - [Text classification](https://halfspacelabs.com/use-cases/text-classification.md): sort text into your own categories - [Personal data redaction](https://halfspacelabs.com/use-cases/personal-data-redaction.md): mask names, emails, phone numbers and addresses in transcripts and tickets - [Intent detection](https://halfspacelabs.com/use-cases/intent-detection.md): name the intent of each customer message, across dozens of intents - [Log triage](https://halfspacelabs.com/use-cases/log-triage.md): label each error log line with its service and whether it should page - [Lead scoring](https://halfspacelabs.com/use-cases/lead-scoring.md): score inbound leads from what they wrote in a form - [Receipt extraction](https://halfspacelabs.com/use-cases/receipt-extraction.md): read merchant, date and total from receipt text - [Content moderation](https://halfspacelabs.com/use-cases/content-moderation.md): moderate posts by your own community rules - [Review tagging](https://halfspacelabs.com/use-cases/review-tagging.md): tag product reviews with sentiment and topics - [Resume parsing](https://halfspacelabs.com/use-cases/resume-parsing.md): turn resume text into clean fields - [Orders from email](https://halfspacelabs.com/use-cases/order-emails.md): turn customer order emails into clean orders - [Facet vs a frontier model](https://halfspacelabs.com/compare/frontier-model.md): when to use each, side by side - [Facet vs fine-tuning](https://halfspacelabs.com/compare/fine-tuning.md): what doing it yourself takes, and when it's the better choice - [FAQ](https://halfspacelabs.com/faq.md): what Facet is, quality, cost, data, ownership, access and security - [Pricing](https://halfspacelabs.com/pricing.md): plans, what a build and a call cost, and Facet next to frontier models at list prices ## Links - [Connect Facet](https://halfspacelabs.com/setup.md): steps any coding agent follows to add the Facet MCP server and sign in - [Facet](https://halfspacelabs.com/index.md): the landing page, also in Markdown - [Facet app](https://facet.halfspacelabs.com): sign in to review, approve and manage your models - [Facet docs](https://halfspacelabs.com/docs): guides and reference, each page also in Markdown (listed under Docs below) - [Changelog](https://halfspacelabs.com/docs/changelog): what's new in Facet, month by month - [Terms of service](https://halfspacelabs.com/terms) - [Privacy policy](https://halfspacelabs.com/privacy) - [Acceptable use policy](https://halfspacelabs.com/acceptable-use): what Facet may not be used for, and how it is checked - [Security](https://halfspacelabs.com/security): what Facet keeps and for how long, who can read it, and how to report a vulnerability - [Subprocessors](https://halfspacelabs.com/subprocessors): the providers that process customer data for Facet, and what each one sees ## Writing Essays, research and notes from the team, each also served as Markdown. - [Why every team will own its model](https://halfspacelabs.com/writing/why-every-team-will-own-its-model.md): why teams will move from renting general models to owning specialized ones, and how agents make that possible ## Docs Every docs page is also Markdown at its path with .md added. - [Overview](https://halfspacelabs.com/docs/index.md): Facet builds a custom AI model for a task your product repeats. Describe the task, and Facet builds, tests, runs and maintains a small model you own. - [Quickstart](https://halfspacelabs.com/docs/quickstart.md): Make your first Model, call it from your code, and ask for it to get better. - [Write good evals](https://halfspacelabs.com/docs/evals.md): Say what a good answer does, in plain statements Facet can check on every change. - [Add examples](https://halfspacelabs.com/docs/examples.md): Upload real inputs and answers as CSV or JSON lines, and let Facet write the rest. - [Make a Model better](https://halfspacelabs.com/docs/improve.md): Say what should change, approve the quote, and Facet makes the fix, checks it and puts it live. - [Return structured output](https://halfspacelabs.com/docs/structured-output.md): Get answers as plain text, as fields you name, or as JSON of any shape. - [Use Facet from an agent](https://halfspacelabs.com/docs/agents.md): Connect any MCP client to Facet, sign in, and run the whole loop from your agent. - [Set up your team](https://halfspacelabs.com/docs/teams.md): Make a shared workspace, invite people, and have your coding agent make Models where the team can see them. - [Export your model](https://halfspacelabs.com/docs/export.md): Download a Model's live version, its Instructions, examples and evals, and run it anywhere. - [Call a Model](https://halfspacelabs.com/docs/call.md): Call any Model through the OpenAI-compatible API, with every parameter and header it takes. - [Errors](https://halfspacelabs.com/docs/errors.md): Every error a Model call or the API returns, why it happens and what to do. - [Limits](https://halfspacelabs.com/docs/limits.md): How many calls, tokens and running Changes a workspace gets, and what happens past them. - [Base models](https://halfspacelabs.com/docs/base-models.md): The open models Facet builds every Model on, what each suits, and what calls cost. - [Pricing and billing](https://halfspacelabs.com/docs/pricing.md): What building, improving and calling a Model costs, how quotes work, and how you're billed. - [The /v1 API](https://halfspacelabs.com/docs/api.md): The HTTP API behind the web app and the MCP tools: authentication, errors and every endpoint. - [Use a Model with the Vercel AI SDK](https://halfspacelabs.com/docs/vercel-ai-sdk.md): Call a Facet Model from the Vercel AI SDK with the OpenAI-compatible provider, in Next.js or any TypeScript app. - [Use a Model with LangChain](https://halfspacelabs.com/docs/langchain.md): Call a Facet Model from LangChain in Python or JavaScript with ChatOpenAI and your Facet serving URL. - [Use a Model with LlamaIndex](https://halfspacelabs.com/docs/llamaindex.md): Call a Facet Model from LlamaIndex with the OpenAILike LLM and your Facet serving URL. - [Use a Model with LiteLLM](https://halfspacelabs.com/docs/litellm.md): Call a Facet Model through LiteLLM, from the Python SDK or the LiteLLM proxy, as an OpenAI-compatible model. - [Use a Model with Pydantic AI](https://halfspacelabs.com/docs/pydantic-ai.md): Call a Facet Model from a Pydantic AI agent with the OpenAI provider and your Facet serving URL. - [Move a call from the Anthropic SDK](https://halfspacelabs.com/docs/from-anthropic.md): Replace one Claude call made with the Anthropic SDK with a Facet Model, using the OpenAI SDK, in Python or TypeScript. - [Concepts](https://halfspacelabs.com/docs/concepts.md): Models, Changes, evals and Versions, and how they fit together. - [Privacy and data](https://halfspacelabs.com/docs/privacy.md): What Facet keeps, who sees your data, how hidden examples stay hidden, and how to delete. - [Changelog](https://halfspacelabs.com/docs/changelog.md): What's new in Facet, month by month. ## Optional - [OpenAPI contract](https://halfspacelabs.com/docs/openapi.yaml): every /v1 operation - [All docs in one file](https://halfspacelabs.com/llms-full.txt) - [Connect Facet to a coding agent](https://halfspacelabs.com/setup.md)