Summary

MiroFish-CLI is a multi-agent AI prediction engine designed for simulating social scenarios. It ingests documents to build a knowledge graph, generates diverse AI agent personas, and simulates their interactions on social media platforms to predict event unfoldings. This tool benefits researchers and analysts by providing machine-readable verdicts and comprehensive reports for agent-driven workflows.

How It Works

The engine processes input documents (PDF, Markdown, text) to construct a knowledge graph and define AI agent profiles. These agents then engage in simulated social media interactions across dual platforms (Twitter and Reddit), posting, replying, and following. A final AI analysis synthesizes simulation data into a predictive report, complete with confidence scores and signals, alongside a machine-readable verdict.json.

Quick Start & Requirements

Requires Python 3.11-3.12 and the uv package manager. Setup involves copying .env.example to .env and configuring the LLM_PROVIDER (defaulting to claude-cli, requiring a Claude Code subscription; codex-cli is also supported). Installation is via uv sync, followed by running simulations with mirofish run --files <source_files> --requirement "<prediction_goal>".

Highlighted Details

Offers both rich CLI visualizations and machine-readable JSON output (--json). Generates detailed run artifacts including ontology.json, graph.json, timeline.json, verdict.json, and various SVG visualizations. Supports parallel, Twitter, and Reddit simulation platforms.

Maintenance & Community

The provided README does not detail specific contributors, sponsorships, or community channels (e.g., Discord, Slack).

Licensing & Compatibility

Licensed under AGPL-3.0. This strong copyleft license necessitates that any derivative works or software linking to this code must also be released under the AGPL, potentially impacting integration into closed-source commercial products.

Limitations & Caveats

The tool is strictly dependent on Anthropic's Claude CLI or OpenAI's Codex CLI for LLM operations, rejecting other providers. Configuration requires setting the LLM_PROVIDER environment variable, and setup involves managing LLM subscriptions.