Argus

Argus is an intelligence layer between employees and the data silos where a company's knowledge actually lives: GitHub, internal docs, the company wiki, Jira, and Slack. Ask a question in plain English and get back a synthesized, source-cited answer in seconds, instead of hopping between tools and pinging coworkers to piece it together yourself.

At its core is an agentic harness I designed that reads each request, decides which sources can answer it, and pulls from the right ones. Then it reasons over what it finds the way a well-informed colleague would, always citing where every claim came from.

And it compounds. Most AI assistants either forget everything between sessions or silently cache their own answers and slowly drift into confident-but-wrong. Argus does neither. It runs a three-tier memory: a trusted layer of human-approved knowledge pages the agent treats as ground truth, a review queue, and the raw exhaust of every past answer. The agent and a background curator can propose changes to trusted knowledge, like opening a pull request, but a human approves before anything becomes trusted. So Argus gets sharper with use, and every change to what it 'knows' carries a name and a timestamp.