These are real agentic applications: multiple specialised agents that reason in several steps, call live external tools, and keep memory across a run. Each one is running right now on our own server — click through and use them.
How these differ from our 100 automations. The automation catalogue is built for speed — one input, one AI pass, one polished output. These agents are a different class of system: they decide which specialist should act, fetch real data from live sources (markets, HackerNews, arXiv, GitHub), verify and score what they find, and build on each other's output before answering.
A Web Agent and a Finance Agent working as a team under a leader. Ask about any stock — the team decides who answers, pulls live prices, analyst recommendations and news from Yahoo Finance, and returns one synthesised report.
A 4-stage pipeline that pulls live signals from GitHub, arXiv, HackerNews, Medium and HuggingFace, scores each one for relevance, assesses risk, then synthesises an actionable intelligence digest.
Three specialists — a HackerNews researcher, a web searcher and an article reader — coordinated as a team. Give it a topic and it researches across sources, opens the actual articles, and writes the report.
Upload a real CV and the team screens it against a role: a technical evaluator scores the fit, a communicator drafts the candidate response, and a scheduler handles next steps.
Every agent must pass a trust verification before it is allowed to participate, and every action is written to a SHA-256 hash-chained audit trail. Tamper with any record and all following hashes visibly break.
Built on the Agno agent framework in Python, deployed as independent services behind nginx. Each agent runs in its own process with its own tool permissions.
Demos call live external APIs, so a run can take 10–60 seconds depending on how many sources the agents decide to consult.