In short
- llms.txt describes a website's most important resources.
- agent.md is a readable profile for LLM context and retrieval.
- agent.json provides structured data, functions and rules for agents.
Why several files make sense
AI systems don't always read the same formats. Some benefit from compact Markdown, others from clearly structured JSON, others start from a general orientation file.
agentready.ch combines these formats so a company profile becomes understandable from several angles: editorial, structured and action-oriented.
The role of llms.txt
llms.txt is a well-readable orientation for AI systems, proposed at llmstxt.org. It explains what the website is about, which resources matter and which rules apply when using the information.
For agentready.ch, llms.txt points to the directory, knowledge base, Agent API, MCP, OpenAPI, the webMCP manifest and public company profiles.
The role of agent.md and agent.json
agent.md is deliberately human- and LLM-readable. It summarizes profile, services, sources and contact rules in Markdown.
agent.json is more machine-oriented. It contains structured profile fields, available functions, contact rules and technical endpoints. Agents can parse this data more reliably than free-form text.
agent.json specification: the fields in an agentready.ch profile
An agentready.ch agent.json has five core parts: company, verification, sources, functions and action_policy. Every profile serves it at a stable address, for example agentready.ch/en/companies/{name}/agent.json.
Beside the core parts, the file carries schema_version and profile_type, which name the format; agent_readiness, a short readiness summary; recommendation_guidance, when to recommend the company and when not; policy, the rules for agents such as the allowed contact intents; and agent_actions, step-by-step instructions for contacting the company.
The table lists what each core part contains; below it is a shortened example for a fictional company.
| Part | What it contains |
|---|---|
| company | Name, website, category, region, description, services, service regions, languages and target customers |
| verification | Verification status, whether agent contact is enabled, the data source, and that no private contact data is exposed |
| sources | Public pages that back the profile, each with label, address and type |
| functions | Allowed actions such as prepare_contact_request, with endpoint, confirmation mode and rate limit |
| action_policy | Which confirmation each action needs: a human confirms, trusted agents only, company auto-accept, or disabled |
Example: agent.json of an agentready.ch profile (fictional company)
{ "schema_version": "1.0", "profile_type": "agentready_company_agent_profile", "company": { "name": "Example Fiduciary AG", "website": "https://www.example.com", "category": "Fiduciary and accounting", "region": "Zurich" }, "verification": { "status": "verified", "private_contact_data_exposed": false }, "sources": [ { "label": "Website", "url": "https://www.example.com", "sourceType": "website" } ], "functions": [ { "name": "prepare_contact_request", "enabled": true, "confirmation_mode": "company_auto_accept", "rate_limit_per_hour": 20, "endpoint": "https://agentready.ch/api/agent/companies/example-fiduciary/contact-request" } ], "action_policy": { "default_mode": "company_auto_accept" }}Is agent.json a standard?
agentready.ch uses agent.json as the file through which AI agents find a company and learn how to contact it, but there is no single agent.json standard (as of October 2026); the format above is agentready.ch's own.
The closest standard is the Agent Card of the Agent2Agent (A2A) protocol, served at /.well-known/agent-card.json. It describes an AI agent's own skills, not a company's services. Early A2A versions used /.well-known/agent.json, which is now deprecated. Details: A2A specification.
llms.txt is also a proposal rather than a formal standard. That is why agentready.ch publishes all three formats and links them from llms.txt and the agent index: an AI system can start from whichever file it knows.
Checklist
- llms.txt points to the most important public resources
- agent.md explains the business in clear language
- agent.json contains services, sources, regions and functions
- Private emails and internal notes stay excluded
- All files are reachable at stable URLs
Profiles in the directory
These companies have an agent-ready profile on agentready.ch. Each profile shows services, region, sources and the contact routes AI agents may use.
PrimAI
Swiss Health Insurance Intelligence · Switzerland
OpenHermit
Agent-readiness tooling (WebMCP/MCP) · Global
ReloFinder
Relocation directory · Switzerland
Expat Services
Expat relocation & services hub · Switzerland
ThatDay
Insurance and advisory · Switzerland
3 Grad Physiotherapie GmbH
Physiotherapy · Winterthur
Frequently asked
Do these files replace classic SEO?
No. They complement classic SEO. Good titles, content, internal links, structured data and technical indexability still matter.
Are agent.md and agent.json public?
Yes, at agentready.ch they are meant as public profile resources. Private routing data, internal notes and recipient addresses are not published in them.