Batoi AIF Docs

Status and Roadmap

Current implementation status and next hardening areas for Batoi AIF.

Status and Roadmap

Batoi AIF has moved beyond a planning-only framework. The public repository now includes the core gateway, contracts, provider implementations, prompt rendering, policy checks, audit logging, RAD adapters, queue/vector contracts, in-memory queue/vector implementations, REST envelopes, documentation, examples, and smoke tests.

Use this page to understand what is available now and what should be treated as the next production hardening track.

Implemented Foundation

  • standalone PHP framework core
  • governed AifGateway execution
  • provider contracts and registry
  • mock provider
  • OpenAI-compatible provider
  • prompt registry contracts
  • prompt versioning and approval-state support
  • prompt renderer
  • static policy engine and policy decision values
  • audit log contract
  • in-memory audit log
  • REST-style {ok,data,error} envelopes
  • inference request and response values
  • embedding request and response values
  • moderation request and response values
  • streaming event value
  • provider capability values
  • RAD execution context resolvers
  • RAD role permission checker
  • queue adapter contract
  • in-memory queue adapter
  • vector store contract
  • in-memory vector store
  • vector record, search request, and search result values
  • Composer install path
  • GitHub ZIP/drop-in autoload path
  • standalone queue/vector example
  • smoke test

Implemented Documentation

The public repository includes documentation for:

  • framework independence
  • adapter rules
  • distribution policy
  • REST API contract
  • RAD persistence profile
  • ZIP installation
  • queue adapters
  • vector store adapters
  • UIF admin contract
  • Laravel adapter design
  • Symfony adapter design

The Batoi website documentation mirrors the public concepts and links developers to the public repository for source-level details.

Baseline Versus Production Adapters

Queue and vector support is implemented at the contract and in-memory baseline level.

This means developers can:

  • validate queue adapter behavior without external infrastructure
  • validate vector-store behavior without an external vector database
  • test payload rules, tenant filters, and retrieval contracts
  • build real infrastructure adapters against stable AIF contracts

Production adapters remain the next hardening track. Examples include:

  • Redis queue
  • RabbitMQ queue
  • AWS SQS queue
  • database-backed queue
  • RAD/MySQL vector baseline
  • pgvector
  • Qdrant
  • Pinecone
  • Weaviate

Framework Adapter Track

Laravel and Symfony adapter documentation exists as implementation guidance. The intended boundary is clear:

  • AIF core must remain usable without Laravel or Symfony installed.
  • Laravel and Symfony adapters should live as optional integration layers.
  • Controllers, jobs, commands, and message handlers must call AifGateway or AifApi.
  • Framework adapters must not call AI providers directly.

The next step is package-level implementation and validation for these optional adapters.

UIF and Operational Surfaces

The UIF admin contract is defined. The next productization step is to build operational surfaces for:

  • provider management
  • prompt registry review
  • policy configuration
  • execution register
  • audit detail views
  • human review queues
  • evaluation results

These should be rendered with Batoi UIF and backed by AIF contracts or RAD integration profiles.

Next Hardening Priorities

Recommended next work:

  • production queue adapters with retry, idempotency, failure audit, and policy recheck support
  • production vector adapters with tenant isolation, metadata filters, access-control filtering, and citation evidence
  • richer provider capability negotiation across text, embedding, moderation, streaming, and structured output
  • evaluation packs for quality, safety, cost, latency, and workflow acceptance
  • RAD persistence implementation and upgrade scripts where AIF is deployed inside RAD applications
  • operational dashboards and review queues using Batoi UIF
  • release packaging notes for Composer, GitHub ZIP, and optional adapter packages

The public repository and release notes remain the source of truth for exact completed capabilities.