EVIDION

Plan-first AI research for people who want sources, not guesses.

Evidion is a free, open-source terminal research assistant. It shows a research plan before it searches, lets you approve or change that plan, then saves a local, source-linked research trail and report.

Why Evidion?

• It shows the research plan before searching.
• It keeps a local session trail of sources, claims, and open gaps.
• It separates source-reported claims from independently verified facts.
• It works from the terminal and can use your own SearXNG instance.
• It exports portable Markdown, text, HTML, and JSON reports.

Install

Evidion supports Python 3.8 or newer on macOS, Windows, and Linux.

python3 -m pip install --user evidion
evidion doctor

Start Evidion from any directory:

evidion

If the command is not found, add Python's user scripts directory to your PATH. Common locations are $HOME/.local/bin on Linux, $HOME/Library/Python/3.9/bin on macOS, and the Python Scripts directory under %APPDATA% on Windows.

Quick start

Web research needs a SearXNG instance with JSON output. Configure it once:

evidion config set --searxng-url http://localhost:8080
evidion research "Are local AI models useful for small companies?" --mode normal

Evidion displays the plan first. Press Enter to approve, write add to add constraints such as “focus on Switzerland”, or write cancel before any web search happens.

What it does today

1. Proposes a research plan before searching.
2. Searches a SearXNG instance you control.
3. Safely reads public HTML; private/local addresses, unsafe redirects, and non-HTML responses are blocked.
4. Saves sources, source-reported statements, possible conflicts, gaps, and the approved plan in a resumable local session.
5. Exports Markdown, text, HTML, and JSON reports.
6. Includes configuration for local and API-backed model providers. Provider workflows are early-stage and may change.

Example output

Research complete
  Sources discovered:  8
  Claims extracted:    21
  Possible conflicts:  2

  Reports saved:
    Markdown  ~/Documents/Evidion/Reports/report.md
    HTML      ~/Documents/Evidion/Reports/report.html
    JSON      ~/Documents/Evidion/Reports/sources.json

Privacy and safety

• Search queries go only to your configured SearXNG instance.
• Web content is treated as untrusted data, never as instructions.
• The reader refuses local/private addresses, unsafe redirects, unsupported protocols, and non-HTML content.
• API keys are not stored in research sessions or reports.

See SECURITY.md and docs/security.md.

Limitations

Evidion does not verify truth automatically. It extracts and compares source-reported claims; human review remains necessary. Search coverage, source quality, and model output depend on your configuration.

Project status

Evidion is early-stage alpha software. It is useful for controlled research experiments, but APIs, provider support, and report formats may change between releases.

Roadmap

• Plan-first CLI workflow
• SearXNG search and safe HTML reader
• Local sessions and Markdown/JSON exports
• Local and API-backed provider configuration
• Additional search providers (for example Tavily or Exa)
• Browser-based extraction
• Source-credibility signals
• Community source scanning

Comparison

Commands at a glance

See the command guide and provider guide.

Development

git clone https://github.com/EvidionAI/evidion.git
cd evidion
python3 -m pip install -e .
python3 -m unittest discover -s tests -v

Older source releases stay available through their Git tags, for example:

python3 -m pip install --user "git+https://github.com/EvidionAI/evidion.git@v1.0.0"

Where files are stored

Reports and sessions default to ~/Documents/Evidion/. Configuration is platform-native: macOS uses ~/Library/Application Support/Evidion, Windows uses %APPDATA%\\Evidion, and Linux uses $XDG_CONFIG_HOME/evidion (or ~/.config/evidion). EVIDION_HOME and EVIDION_CONFIG override these.

Name and independence

Evidion is an independent open-source project and is not affiliated with similarly named commercial products or services.

Contributing

Issues and pull requests are welcome. Please read CONTRIBUTING.md, the Code of Conduct, and the architecture overview.

License

MIT