# AFA Protocol Canonical site: https://afaprotocol.com/ Operator: Infrastructure Driven Growth And Future (IDGAF Holdings LLC) Organization site: https://www.idgaf.world/ AFA Protocol helps teams control what tool-using AI agents may do. On a supported host, it can check an action before the tool runs, pause higher-risk work for a person, and keep a signed record of what happened afterward. The goal is more useful agent work without unlimited access: longer tasks with stop points, spending that can pause for approval, several agents with clear owners, and better records for review. These are practical operating examples. AFA does not claim a measured capacity multiplier, token reduction, or cost saving. Primary product surfaces: - MCP: connects AFA action checks and signed records to compatible agent hosts. - Hosted API: Coming Soon. Team accounts, durable storage, usage counting, notification delivery, and matching verification results remain launch gates. - SDKs and portable .authority bundles are special cases for particular integrations and evidence handoffs. They are not separate product surfaces. How AFA helps agent work: - Give a task a named owner, purpose digest, time mode, and parent. - Narrow the tools, files, network access, and spending the task may reach on supported hosts. - Compare signed envelopes as work branches, making changes to actor, purpose digest, scope, action class, time mode, and ordered plan visible. - Allow a lower-risk action, ask a person about a higher-risk action, or deny an out-of-scope action. - Keep signed summaries of the request and result without storing raw prompts, code, or secrets. - Preserve who delegated work to whom and what happened first for later review. Drift boundary: - AFA surfaces structural drift; it does not infer intent from private reasoning. - The envelope delta reports changes but does not enforce them by itself. - Parent-scope containment and complete per-child path and call-budget enforcement remain open gates. - A supported host must deliver the action to an active AFA pre-tool gate. Broad envelopes still permit broad behavior. Cryptographic accountability: - AFA does not inspect hidden reasoning or claim to detect a hallucinated thought. - Standard signatures help show whether a recorded event changed after signing. - Ordered hash links separately help show where the event belongs in the workflow. Approval before an action is different from approval added afterward. - AFA uses standard cryptographic foundations and preserves recorded order. - Order sensitivity is a protocol property, not a separate cryptographic hardness claim. - Signed chronology and content digests may support an IP dispute, but they do not decide ownership, originality, authorship, infringement, or truth. - Token budgets and paid inference are things AFA may govern. Hosted metering remains open, and AFA has not measured token-cost reduction. Important boundaries: - The website is static. Hosted accounts and REST access are Coming Soon. - The authority record stores structural fields and one-way payload digests, not raw prompts or private reasoning. - A valid signature and a correct place in the timeline are separate checks. - A signed event is evidence that particular bytes were signed. It is not proof that the content is true. - AFA works alongside operating-system permissions, IAM, and sandboxes. It does not replace them. Public pages: - / — product story and value for agent work - /security.html — plain-language security checks and limits - /privacy.html — public-site privacy statement - /coming-soon.html — hosted launch gates