Capabilities
Open standards we author
The agent ecosystem will only standardize on things people actually ship. We ship the products, then publish the specs underneath them.
Agent Protocol (AIPs)
The registry root. Numbered, versioned specs — like BIPs for Bitcoin or EIPs for Ethereum, but for agents. Browse the full catalogue.
AIP-3 — Skills
SKILL.md — portable agent skill manifest, aligned with Anthropic Skills (open standard since Dec 2025; OpenAI Codex adopted).
designkit.sh
AIP-4 — DESIGN.md token format and registry. Portable design tokens authored once and consumed by every agent runtime.
canvakit.sh
AIP-5 — CANVAKIT.md. Templates bound to live data sources. Render once, re-render when tools update.
AIP-6 — Companies
agentcompanies/v1 — file-format spec for AI companies: companies, roles, objectives, operators. Filesystem-first, vendor-neutral.
governance.sh
AIP-7 — agentgovernance/v1. Approvals, hash-chained audit logs, autonomy policies. Third-party verifiable.
agencies.sh
AIP-8 — agentagencies/v1. The operating layer for autonomous AI agencies — services, agreements, deliverables, invoices.
AIP-9 — Operators
agentoperators/v1. Runtime contract for a single canonical operator shell with pluggable profile, skills, tools, memory, governance.
Want to contribute or implement?
All three live in public repos under github.com/agentik. PRs, issues, and questions welcome.
GitHub