Local, open-source AI for patent drafting — run models, skills, and agents entirely on your computer, at no cost, forever.
OpenPatent helps you draft and refine patent work where your invention data belongs: on your machine. The core is MIT licensed, the layout is transparent (you can read how agents and tools are wired), and you can use your own local or self-hosted models plus your own skills and automation — no subscription required for self-hosted use.
Website: openpatent.techtank.com.tr
Prebuilt installers (Windows, Linux, macOS) are published on GitHub:
Building from source is also free and stays fully local-first — see Getting started below.
- Local patent drafting — Keep drafting, claims, and strategy work under your control when you use a local or self-hosted model endpoint.
- Open source — MIT license; inspect, fork, and adapt the code.
- Transparent structure — Core logic, agents, and patent tools live in plain view under
packages/openpatent(see Repository layout). - Free forever (self-hosted) — No fee to run the open-source stack yourself; optional paid services elsewhere do not change that.
- Your models, skills, and agents — Plug in OpenAI-compatible local runtimes (e.g. Ollama, LM Studio) and extend behavior with your own skills and agents on your hardware.
Honest scope: some workflows may call out to optional external APIs (for example patent database or provider APIs) when you configure them. Your LLM and custom extensions can still run entirely locally.
OpenPatent is built on OpenCode — the same agent-first, tool-rich foundation used for local AI workflows. Upstream project: github.com/anomalyco/opencode.
- Patent drafting — Applications, claims, specifications, abstracts, and drawing descriptions from plain-language input.
- Prosecution — Office action responses, amendments, and arguments.
- Consulting — Patentability, FTO, landscape, and validity-style analysis (configure tools and data sources as needed).
- Litigation support — Claim construction, infringement and invalidity-oriented outputs.
- Portfolio management — Docket-style tracking, deadlines, and status summaries.
- Strategic planning — Landscape and claim strategy support.
Primary workflow agents include draft, prosecute, consult, litigate, manage, and strategy. Specialist subagents (for example prior-art, analyst, claims-analyst) are available through task delegation. Patent-oriented tools include patent-search, claim-parser, mpep-lookup, docket-query, document-template, compliance-check, and citation-format.
- Bun 1.3+
- A model endpoint — recommended for zero ongoing API cost: a local OpenAI-compatible server (Ollama, LM Studio, vLLM, etc.). Cloud providers are optional.
git clone https://github.com/erdalbektas/OpenPatent.git
cd OpenPatent
bun install
bun devFrom the repo root, bun dev is the development equivalent of the openpatent CLI. For a standalone binary built locally:
./packages/openpatent/script/build.ts --single
./packages/openpatent/dist/openpatent-<platform>/bin/openpatentLocal-first: Select the local usage from providers menu to point OpenPatent at an OpenAI-compatible base URL (for example Ollama at http://127.0.0.1:11434/v1).
To do that manually, put config in openpatent.json or .openpatent/openpatent.json in your project (or the paths described in CONTRIBUTING.md); the file supports $schema for validation.
Optional cloud providers: Also, you can configure hosted APIs (OpenAI, Anthropic, Google, and others) if you choose — that path is optional and not required for local-only use.
Example (local OpenAI-compatible server — adjust host, model selection, and API key rules for your runtime):
{
"$schema": "https://openpatent.ai/config.json",
"provider": {
"openai": {
"options": {
"apiKey": "ollama",
"baseURL": "http://127.0.0.1:11434/v1"
}
}
}
}Environment variables are also supported for provider credentials where applicable.
MIT
Tech Tank
Email: openpatent@techtank.com.tr