OpenFi Research

OpenFi is an agent native financial research vault.

It gives a user's coding agent a disciplined research system made of markdown skills, artifact contracts, validators, and a local viewer. The user brings Codex, Claude Code, OpenCode, Gemini CLI, or another coding agent. OpenFi provides the repeatable research workflow.

Research Only

OpenFi is a research tool only.

It can:

• research a company
• discover companies related to a theme
• compare companies
• audit report claims
• validate research artifacts
• display research outputs in a local viewer

It cannot:

• place trades
• connect to broker accounts
• connect to bank accounts
• manage real money
• rebalance portfolios
• submit orders
• provide personalized financial advice

First Prototype

The first prototype proves this loop:

Install, configure, research, validate, view, audit, improve

The first two research promises are:

Research this company.
Find companies related to this theme.

Quick Start

Install the Python package, build the viewer, then start the local server:

python -m pip install -e .[test]
cd web
pnpm install
pnpm build
cd ..
openfi doctor
openfi new-run --workflow company-research --strategy quality-growth --subject MSFT
openfi serve

Then write real sourced artifacts into the created run folder, validate it with openfi validate-run <run-folder>, and open the printed local URL.

Demo Only

Bundled example runs are for UI and validator testing only:

openfi create-example-run

This writes to examples-output/runs by default. Do not use example runs as real research output.

Minimal Agent Setup

The intended user experience is:

npx skills add openfi-research

Then the user can tell Claude Code, Codex, OpenCode, or another agent:

Use OpenFi Research to research MSFT.

Canonical test prompt:

Use OpenFi Research to analyze Microsoft.

Expected agent behavior:

• Bootstrap or update OpenFi.
• Read SKILL.md, resolver/RESOLVER.md, the routed workflow, and the routed strategy.
• Run openfi new-run --workflow company-research --strategy quality-growth --subject Microsoft.
• Research from current real sources.
• Write OpenFi artifacts under the returned repo_path.
• Run openfi validate-run <run-folder>.
• Start openfi serve.
• Check the viewer /api/diagnostics and confirm runs_dir matches the run folder parent.
• Check /api/runs and confirm the created run id is visible.
• Report the exact repo path, run folder, validation command, validation status, viewer URL, and blockers.

The test fails if the agent creates custom tooling, invents a separate viewer or validator, writes outside the bootstrap JSON repo_path, copies examples, uses demo fixtures as research, or reports a viewer URL whose diagnostics index a different runs directory.

The installed launcher skill lives at launcher/openfi-research/. It tells the agent to clone or update this repo, install the Python package, build the viewer with pnpm, read the repo skills, create a research run, validate it, and return the viewer URL.

Install the launcher skill into Codex with:

openfi install-skill

Until openfi-research is published to a skill registry, other agent hosts can install by copying launcher/openfi-research into the agent's local skills folder.

How Agents Use OpenFi

Agents start with SKILL.md, then load resolver/RESOLVER.md, then choose the matching workflow skill.

Company research loads workflows/company-research/SKILL.md plus strategies/quality-growth/SKILL.md by default.

Every research run is file backed. The viewer renders those files. It does not own hidden state.

Artifact Shape

Research runs live under artifacts/runs/ or examples/runs/.

Required files:

run.yml
run_card.md
report.md
sources.jsonl
claims.jsonl
assumptions.yml
checks.json

Theme discovery runs also include:

candidates.jsonl

License

See LICENSE.