AI Objective Index

AI Objective Index (AOI) is a read-only MCP/API data contract and vNext AI Agent Capability Trust Router for routing explicit objectives to source-traced capability candidates with constraints, missing fields, route decisions, local receipt memory, and local metadata probes.

Why This Exists

AI agents increasingly need to choose between tools. A generic web search result, sponsored directory, or untraced recommendation is not enough when the agent must explain why one option fits an objective better than another. AOI exists to make those comparisons explicit, auditable, and bounded.

AOI is not a generic web search engine. It is not a supplier-submitted directory. It is not a paid ranking system. It is not a payment, booking, email, login, form submission, purchase, or contract execution platform.

AOI is a productization prototype built from internal research concepts. Internal research claim ceilings do not prohibit implementation; they prevent unsupported external claims about readiness, guarantees, certification, or authority.

Current Scope

Package 0 defines the v0.1 project identity, schemas, documentation, sample data contract, and claim boundaries for the first vertical: AI tools, APIs, SaaS products, and MCP servers.

v0.1 is read-only. Package 1 adds the local core engine, Package 2 adds read-only MCP tool functions, and Package 3 adds a read-only FastAPI REST API. Package 6A adds a crawler/extractor skeleton and fixture pipeline, but it does not run broad live crawling. This repository still does not implement a production ingestion pipeline, payment flow, booking flow, account login, email sender, form submitter, purchasing system, supplier claim/verify flow, profile modification, or contract workflow.

vNext adds conservative routing surfaces: Capability Trust, Objective Router, ExecutionReceipt memory, and local-only Probe-before-Use. These are source-traced route-decision aids, not verification, safety certification, quality guarantee, product readiness, live security scanning, external gateway execution, or action authorization.

AOI-AGENT-ADOPTION-1 reframes AOI 0.3.0a2 for ordinary AI agents as agent-native capability discovery plus pre-use trust routing. Discover mode helps an agent find useful source-traced MCP/tool/API candidates even when every candidate remains HOLD. Preflight mode separates tool_available from tool_authorized, preserves missing fields and must-not-claim boundaries, and routes enterprise or higher-risk cases to ResidualOps without adding live tool calls, external actions, PyPI upload, MCP Registry publish, certification claims, product-readiness claims, or private-kernel disclosure.

AOI-AGENT-ADOPTION-2 wires that agent-native pack into local read-only REST and MCP surfaces before any 0.3.0a2 upload or Registry publish. New REST endpoints expose the capability card, discover mode, preflight mode, and adoption status. New MCP tools expose the same surfaces for agent clients. This keeps the product sentence explicit: discover first, preflight second, claim boundary always. It still performs no PyPI upload, MCP Registry publish, live MCP/tool call, external API call, GitHub API call, token handling, certification claim, product-readiness claim, action authorization, or private-kernel disclosure.

AOI 0.3.0a2 final publish tooling adds the guarded real distribution path for the agent-native package. Final preflight checks version and marker sync, agent REST/MCP surfaces, package-data inclusion, no-secrets status, claim guard, tech protection, and public/private alignment. PyPI upload remains an explicit local action requiring AOI_REAL_PYPI_UPLOAD_CONFIRM=YES and interactive twine token entry. MCP Registry publish remains an explicit local action after real PyPI install verification and AOI_MCP_REGISTRY_SUBMIT_CONFIRM=YES. The tooling does not store tokens, commit .pypirc, commit dis

... View full README on GitHub