biointelligence-agent README

The project's README from github.com/ray-r-ren/biointelligence-agent. As one markdown file: README.md.

Biosecurity Agent Harness v0.1

AI agent that builds a live biosecurity world around any target.

Biosecurity Agent visual world

Quick Start

npx @forsy/biosecurity-agent

Configure your AI agent. Tell it what to protect. It starts investigating and tracking automatically.

Start with one target or a connected set, then refine them conversationally as the world changes. The terminal exposes each real processing lane from discovery and retrieval through claim extraction and synthesis while keeping the underlying evidence inspectable. Exit whenever you like; the local runtime restores the same targets, watchers, and world on your next launch.

1. Define targets

Describe people, animals, plants, products, places, organisations, or multiple connected targets in ordinary language. Files, private context, public URLs, and custom sources can add the details that make evidence relevant to you.

2. Live biosecurity

Your agent tracks official and scientific sources, news and the open web, public OSINT, sensors and surveillance, and connected data. It builds a target-centred biosecurity world, links evidence across places and time, and keeps that world updated in the background.

3. Predict + protect

From the current live state, the agent can simulate forward and recommend defensive actions backed by inspectable evidence. Observed, inferred, and simulated claims always remain distinct.

Terminal

This excerpt comes from the persisted run shown above:

TARGET MODELLING          ✓ 3 targets
OFFICIAL + SCIENTIFIC     ✓ 5 sources
NEWS + OPEN WEB           ✓ 4 sources
SOCIAL + COMMUNITY        ✓ 3 sources
SENSORS + SURVEILLANCE    ✓ 4 sources
CUSTOM SOURCES            ✓ 2 sources
WORLD SYNTHESIS           ✓ 18 entities · 18 relationships
LIVE WATCH                ● 3 watchers

SIMULATION · +14 DAYS
London household · Milo · Heathrow → Singapore
3 future paths · 1 protection · 3 evidence links

Open the full terminal replay.

The terminal remains the controller for targets, sources, evidence, simulations, protection, and optional alerts. Its world, processing events, watchers, and target relationships persist locally and restore on restart.

Visual viewer

Biosecurity Agent is terminal-first; view opens an optional read-only world visualizer for maps, relationships, evidence, and simulations.

Agent compatibility

All agents and model providers are supported through the OpenAI-compatible adapter, with native Codex and Claude adapters.

Privacy + safety

Biosecurity Agent runs locally. Targets, files, and secrets are not uploaded to us by default. External content is treated as untrusted, actions require explicit permissions, and the project is designed for defensive biosecurity, not pathogen engineering or clinical diagnosis.

Dataset

https://huggingface.co/datasets/Forsy-AI/BiosecurityAgent01-SeedDataset

License

Apache-2.0. See LICENSE.

biointelligence-agent (ray-r-ren/biointelligence-agent)

Searching Database…

这是什么?它是一个 AI 智能体框架,会围绕你定义的任意目标构建一个实时演化的「生物安全世界」,持续收集证据并推荐防护行动。
这是图形界面应用还是命令行工具?它以终端为主:你通过 CLI 管理目标、信息源、证据和模拟推演,另有一个可选的只读网页可视化工具,用于查看地图和关系。
怎么让它跑起来?你配置好 AI 智能体,用日常语言告诉它要保护什么;随后它会自动开始调查和跟踪,会话的目标与世界状态会在本地持久保存,重启后依然保留。
它支持哪些 AI 智能体或模型?任何智能体或模型提供商都能通过 OpenAI 兼容适配器接入,另外为 Codex 和 Claude 内置了原生适配器。
这是用于病原体研究或临床诊断的吗?不是。它明确限定于防御性生物安全监测与防护,不用于病原体工程或临床用途。
我的数据会被上传到别处吗?不会。智能体在本地运行,目标、文件和密钥默认不上传;它拉取的任何外部内容都被视为不可信。
我可以跟踪哪些类型的目标?人物、动物、植物、产品、地点、组织,或多个相互关联的目标,全部用日常语言描述。
智能体会监控哪些信息源?官方和科学来源、新闻和开放网络、公开 OSINT、传感器和监控数据流,以及你添加的任何关联数据。
除了监控它还能做什么?能。它可以从当前实时状态出发向前模拟各种情景并推荐防御措施,而且观测到的、推断的和模拟的结论始终明确区分。
Creator
ray-r-ren
Size
1.8 MB
Stars
508
Updated
CommitSep 14
Age
1mo
Commits · 90 Days