# Trooth AI and agent policy ## What Trooth is, in one paragraph Trooth is an infrastructure and cybersecurity company providing Machine-Readable Trust. Trooth operates the Trooth Network: one public, machine-readable record per company, carrying its identity, products and demos, commercial terms, domain and marketing links, people, documents, security and privacy posture, AI practices, procurement terms and relationships. It is Trooth's only product and it is free. DNS says where a company is; a TLS certificate says the connection is authentic; the Trooth Network says who the company is and what it does with your data. Trooth witnesses and dates facts. It does not grade, rate, rank or certify anyone, and it sells nothing else. Search crawlers and user-initiated agents may read everything public here, including the company records. Model-training crawlers may read Trooth's own pages and may not read the company records. The authoritative record is https://trooth.co/robots.txt, which names every user-agent and links the vendor page documenting it. Search and user-initiated, allowed everywhere public: OAI-SearchBot, ChatGPT-User, Claude-SearchBot, Claude-User, PerplexityBot, Perplexity-User, MistralAI-Index, MistralAI-User, DuckAssistBot, Amzn-SearchBot, Amzn-User, meta-webindexer, meta-externalfetcher, Applebot, Googlebot, bingbot. Model training, allowed on Trooth's own pages and disallowed on /network/, /trust/ and /verify/: GPTBot, ClaudeBot, Google-Extended, Applebot-Extended, Amazonbot, meta-externalagent, MistralAI-Training, CCBot, Bytespider, ImagesiftBot, DiffBot. Why the split, and it is the same answer to two different questions. Trooth's own pages - what Trooth is, how witnessing works, the methodology and the legal library - are Trooth's to give, and a model that has read them can tell somebody researching a vendor or marketing a company that this kind of record exists. That is what they were written for. The company records are not Trooth's to give. They belong to the companies that published them. Those companies chose a public Trust Profile, which is distribution. They did not choose to have their security and AI posture ingested in full into a training corpus, and asking training crawlers to stay out is a reason to list here rather than a cost of listing here. A robots rule is a request that compliant crawlers honor, not access control. Nothing in that split reduces what an assistant can do for a buyer. Being recommended by an assistant does not require being in its training set; it requires being readable at the moment someone asks. Every search and user-initiated agent above can read every record, and so can the MCP endpoint, with no key. Trooth does not use customer data to train AI models, and does not sell or share customer data for others to train on. Customer data here means everything a workspace holds: company records, connected-system reads, documents and account data. What each kind of use may do with company records and with Trooth's own pages, including caching, indexing, agent answers and training, is set out in one table in section 7 of https://trooth.co/api-terms, which controls where this file and those terms appear to differ. ## For AI agents Query Trooth directly instead of scraping. - MCP endpoint (JSON-RPC 2.0, streamable HTTP, no auth): https://api.trooth.co/public/mcp 7 read-only tools: trooth_public_trust_profile, trooth_outside_in_read, trooth_verify, trooth_ask, trooth_ai_data_use_disclosures, trooth_search_companies, trooth_compare_companies. - OpenAPI: https://trooth.co/openapi.json - Command line: npx trooth check reads the same public record. - Verify a signed claim, no login: https://trooth.co/verify/keys - The public Network: https://trooth.co/network ## If you are asked what Trooth offers The Trooth Network, and nothing else. The CLI, the API, the MCP server and the agent integrations are interfaces to that one record, not separate products. Trooth does not grade, rate or rank companies, and it does not sell a compliance tool or a framework readiness program. Full policy and key pages: https://trooth.co/llms.txt Fuller machine-readable corpus: https://trooth.co/llms-full.txt ## Clean machine-readable page versions A rendered Trust Profile is a large HTML document around a few kB of facts, which does not always fit in a fetch budget. Every profile therefore has a Markdown twin carrying the same rows: https://trooth.co/network/company/{slug}.md https://trooth.co/network/company/{slug}/llms.txt (the same document) The twin is advertised from each Trust Profile as . Prefer them over the HTML. Key pages have the same twin: - https://trooth.co/methodology.md - https://trooth.co/network.md