Skip to content

ROADMAP.md

Latest commit

 

History

History
106 lines (75 loc) · 4.78 KB

ROADMAP.md

File metadata and controls

106 lines (75 loc) · 4.78 KB
Raw

Ootils Roadmap

This is a living document. Priorities may shift based on community feedback and architectural discoveries.

Current Status: V1 Alpha — agent-native operating substrate

All seven V1 milestones (M1–M7) shipped long ago; the project has moved well past them. The runtime is a working ootils-core service with 70 SQL migrations (ls src/ootils_core/db/migrations/*.sql), the full /v1/* REST surface (canonical count = docs/openapi.json), and an LLM agent tool surface. Delivered since the milestones below: DRP runtime (fair-share inter-site transfers, ADR-028), SSE /v1/stream for agent subscriptions (ADR-027), the proof machine (inventory snapshots + reco→outcome chaining + FVA, ADR-030), scenario-scoped planning-param overlay (ADR-025), reschedule messages + Firm Planned Orders (ADR-026), per-agent API tokens + scopes + kill switch (migration 064, ADR-029), and location aliases (ADR-031).

Sequencing is no longer driven by this file. The living, dated plan (H0–H4 phasing, the "designed by AI, for AI" bet, per-chantier acceptance criteria) lives in docs/ROADMAP-AGENTS-2026-H2.md and epic #397. The milestone list below is kept for historical record of the V1 thesis.

We are still building in public — architectural decisions go through ADRs (see docs/INDEX.md for the ADR map).

V1 — Core Engine (Shipping)

Goal: Prove the architectural thesis. A minimal but technically rigorous engine that demonstrates graph-based, incremental, explainable supply chain planning — consumable by an AI agent.

Milestones

  • M1 — Data Model

    • SQL schema: nodes, edges, events, scenarios, explanations (M1 shipped the initial schema; the migration set has since grown to 70)
    • Node/edge type registry (docs/node-dictionary.md, docs/edge-dictionary.md)
    • Ingestion from flat files (CSV/JSON) via /v1/ingest/*

  • M2 — Core Engine

    • Temporal Bridge (engine/kernel/temporal/bridge.py)
    • Projected inventory calculation (engine/kernel/calc/projection.py)
    • Incremental propagation, dirty-flag + subgraph expansion (engine/orchestration/propagator.py)
    • Allocation engine (engine/kernel/allocation/engine.py)

  • M3 — Explainability

    • Root cause chain generation (engine/kernel/explanation/builder.py)
    • Structured explanation storage (explanations and causal_steps tables)
    • Explanation API endpoint (/v1/explain/*)

  • M4 — Shortage Detection

    • Shortage node generation (engine/kernel/shortage/detector.py)
    • Severity scoring (qty × days × cost proxy)
    • Shortage → explanation linkage

  • M5 — Scenarios

    • Override mechanism via scenario_overrides table
    • Simulation endpoint (/v1/simulate)
    • Baseline vs scenario diff (/v1/scenarios/diff)

  • M6 — API

    • REST API: ~50 endpoints (/v1/events, /v1/projection, /v1/issues, /v1/explain, /v1/simulate, /v1/graph, /v1/ingest, …)
    • OpenAPI spec (docs/openapi.json)
    • Bearer-token auth with hmac.compare_digest (api/auth.py)

  • M7 — AI Agent Demo

    • Three agent tools: get_active_issues, simulate_override, trigger_recalculation (tools/agent_tools.py)
    • Phase 1 end-to-end demo (demo/phase1.py, tests/integration/test_phase1_e2e.py)

V1 Out of Scope

  • UI (any UI)
  • Multi-echelon planning
  • Capacity constraints
  • Substitution rules
  • Optimization (MILP/LP)
  • Production-grade auth
  • Cloud hosting

V2 — Planning Intelligence (2027)

  • Multi-echelon support (BOM explosion, multi-level netting)
  • Capacity constraints (CapacityBucket integration)
  • Substitution rules
  • Fair-share allocation
  • Full pegging (complete supply→demand linkage)
  • Scenario versioning (Dolt integration pilot)
  • Performance optimization (graph indexing at scale)
  • WebSocket/SSE streaming for real-time agent subscriptions

V3 — AI-Native Operations (2028+)

  • Native AI agent SDK (Python + REST)
  • Autonomous decision execution with guardrails
  • LLM-readable explanation format (natural language wrapper on structured explanations)
  • Multi-agent orchestration primitives
  • Self-healing plan suggestions
  • Continuous learning from planner overrides

How to Influence the Roadmap

Open a GitHub Discussion tagged Ideas or Architecture and make your case.

Decisions are made based on:

  1. Architectural soundness
  2. Community consensus
  3. Founder's 20+ years of operational experience

Not based on: feature requests without justification, vendor pressure, or hype cycles.

Last updated: 2026-07-07 — status figures corrected (70 migrations; DRP / stream / proof machine / aliases shipped); live sequencing delegated to docs/ROADMAP-AGENTS-2026-H2.md. The V1 milestone list below is historical.