OpenPlanter
github.com/shinmegamiboson/openplanter- Category
- AI Agents
- Rank
- No. 450Tools index
- Pricing
- Open Source
- Type
- APP
- Use case
- Data, Retrieval & Knowledge
- Interfaces
- Desktop · CLI
- Builder
- @ShinMegamiBoson
- GitHub
- 2.7k stars
- Latest release
- v0.1.1
- Date
About
A recursive AI investigation agent that analyzes corporate registries, campaign finance records, lobbying disclosures, and government contracts to uncover hidden connections between entities. Features autonomous operation with file I/O, shell execution, web search, and an interactive knowledge graph visualization.
What it does
OpenPlanter turns a folder of source material into a continuing investigation. You give it an objective, then follow its research, tool activity, written findings, and evolving entity map through either a desktop workspace or terminal session.
Why it's ranked here
The project combines a usable investigation interface with unusually broad execution machinery. Persistent sessions, multiple model providers, headless runs, recursive task splitting, and source-linked graph navigation make it credible for substantial research. Its autonomy also raises the burden of workspace isolation and human verification.
What's good
Investigations can move between interactive and automated workflows without changing tools. The desktop app exposes reasoning, findings, credentials, source documents, and graph relationships together. The terminal agent supports resumable sessions, provider switching, local models, bounded recursion, command timeouts, and optional acceptance checks for delegated work.
Tradeoffs
The agent can edit files and execute shell commands inside its workspace, so operators must choose that workspace carefully. Web search requires an Exa key, while hosted models require provider credentials. Local Ollama support removes that credential requirement, but the repository warns that initial model loading may be slow. Findings still require human validation.
How to use it well
Use it for analysts who already have datasets, a concrete research question, and a defensible verification process. Start in an isolated workspace, define narrow objectives and acceptance criteria, inspect source documents behind graph links, then export findings for review. It does not replace authoritative records, fact checking, or investigative judgment.
Technical notes+
The Python package is configured in pyproject.toml for Python 3.10+ with a console entry point and setuptools packaging. agent/engine.py implements recursive execution, parallel subtasks, observation clipping, repeated-command blocking, cancellation, and optional lightweight result judging; notably, judge failures fall back to PASS. agent/tools.py confines resolved file paths to the configured workspace, caps outputs, blocks heredocs and interactive terminal programs, tracks parallel write conflicts, and cleans up background jobs. agent/model.py implements provider communication with urllib, including JSON requests, SSE accumulation, retries, and separate first-byte and stream timeouts. agent/config.py centralizes provider URLs, credentials, limits, and environment configuration. agent/tui.py supplies the Rich and prompt_toolkit terminal experience, while agent/demo.py censors workspace path segments only at rendering time. README.md describes the separate Tauri 2 desktop application, TypeScript/Vite frontend, Cytoscape.js graph, and Playwright, frontend, backend, unit, and integration test surfaces.
Observed
- License
- MIT
- Languages
- Python CLI agent; Rust backend and TypeScript/Vite frontend for the desktop application
- Install surface
- Editable Python installation through pip, Docker Compose, or pre-built macOS DMG, Windows MSI, and Linux AppImage packages
- Interfaces
- Desktop GUI, interactive terminal UI, plain REPL, and headless single-task CLI
- Platform support
- Pre-built downloads are documented for macOS, Windows, and Linux
- Model providers
- OpenAI, Anthropic, OpenRouter, Cerebras, and local Ollama are supported
- Testing structure
- Repository structure includes unit and integration tests, frontend tests, Playwright end-to-end tests, and Rust backend tests
Read from README.md, pyproject.toml, record_demo.py, cast_to_video.py, quickstart_investigation.py, agent/tui.py, agent/demo.py, agent/model.py, agent/tools.py, agent/config.py, agent/engine.py.
What it can do
Analyze corporate registry data to identify entity relationships
Corporate registry datasets → Entity relationship mappings and connections
Process campaign finance records to trace money flows
Campaign finance data files → Financial connection analysis and donor networks
Parse lobbying disclosure documents for influence patterns
Lobbying disclosure records → Lobbying relationship graphs and influence maps
Extract connections from government contract databases
Government contract records → Contractor relationship networks and award patterns
Generate interactive knowledge graph visualizations
Analyzed entity relationship data → Interactive visual network graphs
Perform autonomous web searches for additional entity information
Entity names and search parameters → Supplemental data and verification sources
Execute recursive investigation queries across multiple datasets
Investigation parameters and target entities → Multi-layered connection analysis reports
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.