llms.txt
# Overmind
> Turn your agents’ production data into specialized models you own. Prepare data, train, evaluate, and deploy from one platform.
Overmind turns an AI team's production agent traces into small open-weight models the team owns: trace, curate datasets, evaluate, optimize, train, and serve from one platform. Agents and scripts call it over REST at https://api.overmindlab.ai (header `X-Api-Key: ovr_...` or `Authorization: Bearer ovr_...`; OpenAPI at https://www.overmindlab.ai/openapi.json); traces arrive through the `overmind` Python SDK (PyPI) or any OpenTelemetry exporter. Trained models are served OpenAI-compatible at https://api.overmindlab.ai/api/v1. The platform is open source under AGPL-3.0 (https://github.com/overmind-core/overmind) and can be self-hosted with Docker Compose, or used hosted at https://console.overmindlab.ai.
## When to use Overmind
- [Train and fine-tune small open-weight models on production traces](https://www.overmindlab.ai/product/model-training): Best fit when an AI team runs agents in production and wants a specialized model it owns instead of renting a frontier API. Training is automated from the agent's own traces, benchmarked against the baseline, and served on a hosted endpoint.
- [Build evals from real traffic and optimize agent code](https://www.overmindlab.ai/product/agent-testing): Use when you need regression tests for an agent: eval runs score prompt, code and model changes against a baseline on real traces, and the Optimizer returns the winning change as a scored git diff you apply.
- [Curate training datasets from traces](https://www.overmindlab.ai/product/data-workshop): Use when raw telemetry or uploaded files must become a versioned, audited dataset for evaluation or supervised fine-tuning.
- [Trace and observe agents over OTLP](https://www.overmindlab.ai/product/observability): Use when you want every agent run traced and scored as it arrives, mapped to the capabilities, prompts and tools scanned from the agent's code.
- [Serve trained and frontier models through one OpenAI-compatible endpoint](https://docs.overmindlab.ai/models/inference.md): Point any OpenAI SDK at https://api.overmindlab.ai/api/v1 with an ovr_ API key; trained ft-... models and frontier models share the same gateway.
- [How an agent should call Overmind](https://docs.overmindlab.ai/platform/api.md): REST base URL https://api.overmindlab.ai; authenticate with the X-Api-Key: ovr_... header (or Authorization: Bearer); JWT via POST /api/auth/token/; machine-readable spec at https://www.overmindlab.ai/openapi.json.
## Product
- [Home](https://www.overmindlab.ai/): Turn your agents’ production data into specialized models you own. Prepare data, train, evaluate, and deploy from one platform.
- [Agent testing](https://www.overmindlab.ai/product/agent-testing): Generate evaluations from your agent’s context, compare models and prompts on the same data, and test code changes in your own repository. Inspect the scores, evidence and winning diffs before deciding what to ship.
- [Data workshop](https://www.overmindlab.ai/product/data-workshop): Turn production traces and uploaded files into versioned datasets for evaluation and training. A dedicated dataset agent prepares the rows, checks their quality, and records each transformation as an executable Python cell.
- [MCP](https://www.overmindlab.ai/product/mcp): Bring Overmind into the coding agent you already use. Investigate failures, build datasets, run evaluations and train specialized models — with your repository and production evidence in the same conversation.
- [Model training](https://www.overmindlab.ai/product/model-training): Fine-tune a model on your own data, measure it with your own evaluators, and serve it through an OpenAI-compatible endpoint. Keep the training data, benchmark and deployment connected from the first experiment to production.
- [Observability](https://www.overmindlab.ai/product/observability): Connect what your agent is built to do with what happens in production. Map capabilities from your code, follow each execution, and investigate failures with the prompts, tools, scores and source behind them.
- [Model Library](https://www.overmindlab.ai/models): Open-weight models Overmind trains on your agent's own data, all trainable with LoRA or full finetuning.
- [Pricing](https://www.overmindlab.ai/pricing): Plans for tracing, testing, training, and deployment at any scale.
## Comparisons
- [All comparisons](https://www.overmindlab.ai/compare): How Overmind differs from observability, tracing, and fine-tuning tools, including the cases where a competitor is the better choice.
- [Overmind vs LangSmith](https://www.overmindlab.ai/compare/overmind-vs-langsmith): The most adopted agent observability platform, now with LoRA fine-tuning via smithtune on Fireworks or Baseten. Overmind adds trace import from any source, audited datasets, a benchmark against production, hosted serving and downloadable weights.
- [Overmind vs Langfuse](https://www.overmindlab.ai/compare/overmind-vs-langfuse): The open-source observability standard. Overmind imports its traces today and runs the training loop it deliberately leaves out.
- [Overmind vs OpenPipe](https://www.overmindlab.ai/compare/overmind-vs-openpipe): Proved that production requests should train cheaper models you own. Sunset July 2026; Overmind runs that idea as a full loop.
- [Alternatives to OpenAI fine-tuning, and how to migrate before 6 January 2027](https://www.overmindlab.ai/compare/openai-fine-tuning-alternatives): OpenAI closes self-serve fine-tuning to new jobs on 6 January 2027. The dates, four alternatives, and the migration steps.
## Developers
- [API reference](https://docs.overmindlab.ai/platform/api.md): Base URL, authentication, and the endpoint map for the Overmind REST API.
- [OpenAPI spec (JSON)](https://www.overmindlab.ai/openapi.json): OpenAPI 3 description of the Overmind platform API for code generation and function calling.
- [OpenAPI spec (YAML)](https://www.overmindlab.ai/api/openapi.yaml)
- [Authentication](https://docs.overmindlab.ai/platform/api.md#authentication): API keys (X-Api-Key or Bearer), JWT tokens, and console sessions.
- [Quickstart](https://docs.overmindlab.ai/quickstart.md): Create an API key and send the first traces in minutes.
- [Python SDK](https://docs.overmindlab.ai/tracing/sdk-python.md): The `overmind` package on PyPI: init(), auto-instrumentation, span decorators.
- [OpenTelemetry ingest](https://docs.overmindlab.ai/core/observability.md): Any OpenTelemetry exporter can POST OTLP protobuf to /api/v1/traces. There is no TypeScript SDK; Node.js agents use OpenTelemetry directly.
- [Source code](https://github.com/overmind-core/overmind): The Overmind platform, open source under AGPL-3.0; the SDK and CLI are MIT.
- [Self-hosting](https://docs.overmindlab.ai/platform/self-hosting.md): Run the platform on your own infrastructure with Docker Compose.
- [Inference API](https://docs.overmindlab.ai/models/inference.md): OpenAI-compatible chat completions at https://api.overmindlab.ai/api/v1.
- [Developer console](https://console.overmindlab.ai): Sign in to manage projects, API keys, traces, datasets, and training runs.
- [API catalog](https://www.overmindlab.ai/.well-known/api-catalog): RFC 9727 linkset pointing at the spec and docs.
## Documentation
- [Docs](https://docs.overmindlab.ai/): Product documentation, guides, and API reference (see also https://docs.overmindlab.ai/llms.txt).
- [Full site content](https://www.overmindlab.ai/llms-full.txt): Every page listed here, in the same order, joined into one Markdown file.
- [Full docs content](https://docs.overmindlab.ai/llms-full.txt): The complete product documentation as one Markdown file.
## Glossary
- [All glossary terms](https://www.overmindlab.ai/glossary): Plain definitions of LLM observability, evaluation and training terms, with the questions buyers ask answered in the text.
- [What is LLM observability?](https://www.overmindlab.ai/glossary/llm-observability): LLM observability captures every prompt, completion and tool call as traces so you can explain what your AI did. It cannot tell you whether it was any good.
## Research
- [My summer at Overmind](https://www.overmindlab.ai/research/my-summer-at-overmind): A reflection on learning, autonomy, and building across product surfaces during a summer internship at Overmind.
- [Small is the new frontier](https://www.overmindlab.ai/research/small-is-the-new-frontier): Why specialized models trained on your own production data can beat frontier generalists on economics, ownership, and reliability.
- [When bigger isn’t better](https://www.overmindlab.ai/research/when-bigger-isnt-better): On specialized tasks, Overmind-trained small language models outperform frontier models on accuracy, hallucination, and cost.
- [How to train your agent](https://www.overmindlab.ai/research/how-to-train-your-agent): Seven patterns from teams running AI agents in production: instrument every trace, build evals from real traffic, and label before you collect.
- [Unwrapping the wrappers](https://www.overmindlab.ai/research/who-trains-their-own-models): Cursor, Intercom, Replit, and Perplexity all trained specialist models on their own production data. What each one built, how, and what it bought them.
- [How do you turn traces into a training dataset?](https://www.overmindlab.ai/research/traces-to-training-dataset): The four stages from agent telemetry to a training dataset: capture every run as a trace, sample what to label, score it, and build for SFT or RL.
- [Prompt engineering vs fine-tuning](https://www.overmindlab.ai/research/prompt-engineering-vs-fine-tuning): Prompting changes the input. Fine-tuning changes the weights. RAG changes what the model knows. Side-by-side table, worked examples and when to switch.
- [What are the different types of fine-tuning?](https://www.overmindlab.ai/research/types-of-fine-tuning): The four fine-tuning options and how to pick: supervised fine-tuning, reinforcement fine-tuning with GRPO, distillation, and LoRA or QLoRA to run them cheaply.
- [Open-weight vs frontier models: when to rent, when to own](https://www.overmindlab.ai/research/open-weights-vs-frontier-apis): Open-weight models vs frontier APIs on cost, latency, data residency and lock-in. A side-by-side table, the cost crossover, and when to pick each.
- [The anatomy of an AI agent](https://www.overmindlab.ai/research/anatomy-of-an-ai-agent): What an AI agent is made of: the model, the agentic loop, tools, and memory, plus the trajectory they produce and why it's the unit of work.
- [So, you have observability. Now what?](https://www.overmindlab.ai/research/so-you-have-observability-now-what): You’ve instrumented prompts, traces, and tool calls. Here’s why observability can’t judge quality - and how to close the loop from traces to better models.
- [Secret Agent Overmind](https://www.overmindlab.ai/research/secret-agent-overmind): Diagnose and optimise production AI agents by fusing code, traces, evals, and datasets so every behaviour is explainable, sourced, and fixable.
- [Think Smaller](https://www.overmindlab.ai/research/think-smaller): Specialised small language models, fine-tuned on production data, beat frontier LLMs on most agent tasks. The case for SLMs and owning your weights.
## Company
- [About](https://www.overmindlab.ai/about): Who builds Overmind and why.
- [Contact](https://www.overmindlab.ai/contact): Email, booking a call, support, and company address.
- [Careers](https://www.overmindlab.ai/careers): Open engineering and research roles.
- [Privacy policy](https://www.overmindlab.ai/legals/privacy)
- [Terms](https://www.overmindlab.ai/legals/terms)
## Optional
- [Sitemap](https://www.overmindlab.ai/sitemap.xml)
- [Research RSS feed](https://www.overmindlab.ai/research/feed.xml)
- [Docs llms.txt](https://docs.overmindlab.ai/llms.txt)
- [Markdown versions of every page](https://www.overmindlab.ai/index.md): Append .md to any page path (for example https://www.overmindlab.ai/about.md) or send Accept: text/markdown.
