FINRUN 0.5.0-preview.2 · PUBLISHED
Your question.
One Finrun skill.
Designed for Codex running locally, with Python 3.10+ and a writable workspace. This edition is not supported in ChatGPT web. The public directory link opens the listing; use Finrun from a local Codex task.
Get started
- Open the public listing and install Finrun for Codex.
- Start a local task in a writable workspace. Select the actual @Finrun plugin mention.
- Ask your question. Specify the period, language or format when it matters.
- Review the answer and open any generated report or data sheet. Inspect source dates, calculations and unresolved gaps.
Typing “@Finrun” as plain text is not necessarily the same as selecting the plugin in Codex. If you already have an older personal copy installed, remove that copy before installing the public version to avoid duplicate entries.
What you can ask
“Analyze NVIDIA’s main business and financial risks.”
“Compare the latest earnings of Microsoft and Alphabet.”
“Find three financial developments worth researching.”
Choose the output
Brief questions get concise sourced answers. Substantial analysis can produce an interactive HTML report and structured JSON. Data sheets use CSV by default; request Excel explicitly when you need an XLSX workbook. For topic discovery, Finrun uses its Radar method. It can stop at a shortlist or research the topics you choose. You always select the same Finrun skill.
What you need
- Local Codex, Python 3.10 or newer and a writable workspace.
- Browsing or source-retrieval tools and authorized access to the requested sources.
- Internet access to GitHub for the public method catalog. A verified cached method can be used when fresh retrieval fails, with the limitation disclosed.
- Your own agent access and any required data subscriptions. Yahoo market data uses an optional local dependency.
How it works
One Finrun skill selects relevant financial methods from a public GitHub catalog. Each request records the selected version. Local Python helpers support evidence records, calculations, validation and portable reports. Generated reports and caches stay in the task’s workspace. The agent and source providers have their own data practices.
No separate Finrun account is needed. This edition does not include a hosted MCP service, trading or an automatically enabled schedule. This website displays saved examples; it does not run research.