Every agent in the corpus
A field guide to the systems reshaping how work gets done. Explore the agents and open-source harnesses in the Fieldbook, each classified by deployment mode, inferred topology, home country and build approach. Every specimen card makes the operating model behind the product visible.
The full catalogue: classified, specced and sourced
The catalogue is built to show what an agent is, not just what it promises. Each entry is classified, specified and sourced from available evidence. Its topology is inferred from product descriptions, technical documentation and how the system is configured in practice.
Show more
Vertical SaaS is the leading deployment path at 31 of 103 agents, with security, fraud and finance advancing behind it. In tightly governed enterprise settings co-pilot patterns remain common: they keep information within the business, preserve approval controls and can orchestrate across multiple models. In banking and finance that often means pairing frontier models with custom-trained systems. Filter by modality and classification, then open a card for the full spec sheet.
What are the main types of AI agent?
Broadly five. Frontier assistants from the large labs, copilots built into a tool you already use, platforms for building your own, open-source frameworks you host yourself, and vertical agents made for one industry such as law or healthcare.
What is the difference between an AI agent and an AI assistant?
How much happens without you. An assistant replies to what you ask. An agent takes a goal, plans the steps and uses tools until the task is done. The labels are used loosely in marketing, so check what a product actually does on its own.
How do I choose an AI agent for my business?
Start with the job, not the model. Check what tools and data it can reach, whether it sits inside software your team already uses, what it costs per user or per task, and who approves its actions. Tool access increasingly runs through MCP, an open standard worth checking support for (Model Context Protocol).
What model do most AI agents run on?
Usually a foundation model from one of the big labs rather than one trained in-house. The agent is everything around it: the tools, the memory, the workflow and the guardrails. That is why two agents can behave very differently while sharing the same model.
Can multiple AI agents work together?
Yes. Multi-agent setups split a job across specialised agents, often with one coordinating the rest. It adds capability, and also adds cost, latency and new ways to fail, so most production systems keep the number of agents small.