# Ratchet docs > Ratchet is an OpenAI-compatible gateway that distills open teacher models into a small student. You end up with an adapter you own, learned only from open teachers, with a gated nightly release. The full text of every page is in [llms-full.txt](https://docs.ratchet.dualmindlab.tech/llms-full.txt). ## Start here - [What is Ratchet](https://docs.ratchet.dualmindlab.tech/guides/what-is-ratchet/): A two-minute explanation of Ratchet, an OpenAI-compatible gateway that moves narrow, repeated work from a rented teacher model to a small student (NVIDIA's open Nano base plus an adapter you own), and ships each new student only if it passes a statistical gate. - [Why it matters](https://docs.ratchet.dualmindlab.tech/guides/why-it-matters/): What Ratchet is worth to a team paying for LLM calls, a team that must know where its data went, a regulated team and an open-source builder, with the page that backs each claim. - [The business case](https://docs.ratchet.dualmindlab.tech/guides/business/): Who Ratchet is for, what it saves, how it could earn money, why now, how it differs from other categories of tool, and what is not proven yet. - [A text-to-SQL adapter you can check](https://docs.ratchet.dualmindlab.tech/guides/text-to-sql/): A small model that turns a question and a database schema into one SQLite query, trained only on queries that were run, and scored on whole databases it never saw. - [Results](https://docs.ratchet.dualmindlab.tech/guides/results/): What the text-to-SQL adapters (the headline) and the tool-calling adapter scored on exams they never trained on, and how to read the numbers. - [Quickstart](https://docs.ratchet.dualmindlab.tech/guides/quickstart/): Run the Ratchet gateway on your own machine against Nebius Token Factory, send it a request with the OpenAI SDK, and read the ratchet block in the answer. - [For judges](https://docs.ratchet.dualmindlab.tech/guides/for-judges/): The judges pack: the pitch, a five-minute path, the proof with links and one command per claim, every link in one place, the hackathon checklist, what is real and what is not yet, and what it cost. ## How it works - [The big picture](https://docs.ratchet.dualmindlab.tech/guides/big-picture/): Every Ratchet component, where it runs, how the pieces connect, the live runtime, and the three loops that run at different speeds. - [One request's journey](https://docs.ratchet.dualmindlab.tech/guides/request-journey/): What happens to a single chat request, from the gateway's admission checks through the student, the teacher and the ledger, to the response. - [Confidence](https://docs.ratchet.dualmindlab.tech/guides/confidence/): How the gateway decides whether the student is sure enough to answer, and how each version's threshold is set and checked on held-out golden rows. - [Escalation and audits](https://docs.ratchet.dualmindlab.tech/guides/escalation-and-audits/): When a request goes to the teacher, what the teacher's answer is used for, and how random audits check the student and drive the release gate. - [The ledger](https://docs.ratchet.dualmindlab.tech/guides/ledger/): What Ratchet keeps in its SQLite ledger and what it never keeps, how PII is redacted, and how opt-outs, retention, erasure and export work. - [Night School](https://docs.ratchet.dualmindlab.tech/guides/night-school/): How Ratchet trains a new student every night - the checks, the golden set, the FREQUENT rule, LoRA settings, and what is left out of training. - [The release gate](https://docs.ratchet.dualmindlab.tech/guides/release-gate/): How a newly trained adapter goes from candidate to live through the golden gate, shadow and canary, and how a bad one is rolled back. - [Training on your data with the student endpoint](https://docs.ratchet.dualmindlab.tech/guides/training-on-your-data/): How your traffic becomes a LoRA adapter you own - what the gateway stores, the nightly Nebius job, the bucket, the endpoint, and what is ready to run. - [Licenses: why only open teachers](https://docs.ratchet.dualmindlab.tech/guides/licenses/): Why Ratchet learns only from teachers whose license allows it, how closed models are refused in code, and which licenses travel with the adapter. - [Cost and savings](https://docs.ratchet.dualmindlab.tech/guides/cost-and-savings/): How Ratchet prices each request, counts the GPU-hours you actually pay for, and says plainly when self-hosting the student does not pay. - [Evidence you can check](https://docs.ratchet.dualmindlab.tech/guides/evidence/): Signed bills of materials, certified precision, committed training sets, ledger replay and savings receipts, and what each one does and does not prove. - [Proof you can run](https://docs.ratchet.dualmindlab.tech/guides/proof/): Every claim Ratchet makes comes with a command that checks it, what it proves, and what a failure looks like. - [Check it yourself](https://docs.ratchet.dualmindlab.tech/reference/verify/): Every claim Ratchet makes about a result can be rechecked from files in this repository. This page lists each claim, the command that checks it, what... ## Scout - [Scout](https://docs.ratchet.dualmindlab.tech/guides/scout/): Five research agents whose calls all go through Ratchet, showing which calls the student takes over and which stay pinned to the teacher. - [Scout: the multi-agent research tool](https://docs.ratchet.dualmindlab.tech/reference/scout/): Scout is a small team of five research agents (planner, searcher, triage, reader and writer) that answers a question from the web. Every agent is a... ## Run it - [Where everything runs](https://docs.ratchet.dualmindlab.tech/reference/where-it-runs/): This page says, for every part of Ratchet, where it runs today and where it is designed to run on Nebius AI Cloud. Read the last column first: Real... - [Host it on Nebius](https://docs.ratchet.dualmindlab.tech/guides/host-on-nebius/): Every Nebius piece Ratchet uses and how they connect, then the steps to run the VM, the GPU endpoint and training, with costs. - [Deployment](https://docs.ratchet.dualmindlab.tech/reference/deployment/): There are two ways to run Ratchet: mock (local, no keys, about a minute) and live (real Nemotron on Nebius). The nebius ai commands in deploy/ follow... - [An alternative host: the Nebius CPU VM kit (ran once by hand; the one-command kit is tested offline)](https://docs.ratchet.dualmindlab.tech/reference/deploy-nebius-vm/): This guide puts the live Ratchet gateway on one small Nebius AI Cloud virtual machine, so evaluators can use it if you choose this host. The VM runs... - [Runbook](https://docs.ratchet.dualmindlab.tech/reference/runbook/): How to operate Ratchet in live mode: what to check every day, what the events mean, how to intervene by hand, and what to do when something breaks. The... ## Security - [Security and safety](https://docs.ratchet.dualmindlab.tech/guides/security/): Who may call what, how spend is capped, what is logged and stored, how the Nebius VM kit is locked down, and the rules the learning loop never breaks. - [Threat model: security and privacy](https://docs.ratchet.dualmindlab.tech/reference/threat-model/): Scope: Ratchet v0.5.0 (released and tagged Oct 25, 2026) in live mode, made up of the gateway, the SQLite ledger, the Night School driver, the Nebius... ## API and reference - [HTTP API](https://docs.ratchet.dualmindlab.tech/reference/api/): The gateway (ratchet/gateway.py, FastAPI app ratchet.gateway:app) speaks the OpenAI chat-completions format, so any OpenAI SDK works. It also exposes a... - [Endpoint map](https://docs.ratchet.dualmindlab.tech/api/endpoints/): Every route of the Ratchet gateway in one place, with who may call it, what it is for and a link to its generated reference, plus three copy-paste examples. - [Environment variables](https://docs.ratchet.dualmindlab.tech/api/environment/): Every environment variable Ratchet and its deploy kit read, grouped, with its default, where it matters and what it does. - [Configuration](https://docs.ratchet.dualmindlab.tech/reference/configuration/): All settings come from environment variables. Defaults live in ratchet/config.py (Settings), ratchet/live.py, the deploy/ scripts and deploy/vm/. Model... - [Command line](https://docs.ratchet.dualmindlab.tech/api/cli/): Every Ratchet command and Make target, with when to use it and its help text captured from the code. - [Python API](https://docs.ratchet.dualmindlab.tech/api/python/): Embed Ratchet in a Python program without HTTP. Build the settings, ledger and providers, call Ratchet.handle, and tick the release gate yourself. - [Architecture](https://docs.ratchet.dualmindlab.tech/reference/architecture/): Ratchet is a gateway that sits between your app and a large "teacher" model. A small "student" model, NVIDIA's open Nano base with a LoRA adapter you... - [Testing](https://docs.ratchet.dualmindlab.tech/reference/testing/): Tests without a table of their own yet: test_model_routing, test_attribution, test_exam_numbers, test_reasoning, test_toolcall_results,... ## Help - [Glossary](https://docs.ratchet.dualmindlab.tech/guides/glossary/): Every Ratchet term in one line, in alphabetical order, each with a link to the page that explains it. Underlined words across these docs show these definitions on hover. - [FAQ](https://docs.ratchet.dualmindlab.tech/guides/faq/): Short, honest answers to the questions people ask most about Ratchet, grouped for everyone, evaluators, engineers and buyers. ## HTTP API - [HTTP API reference](https://docs.ratchet.dualmindlab.tech/api/http/): every route, generated from the gateway's OpenAPI schema