Pranor v2.0 — AI Execution Fabric Architecture

Release Status: v2.0.0 GA — Officially merged to main and tagged v2.0.0.

Overview

Pranor v2.0 introduces a governed AI agent execution layer on top of the v1.x infrastructure. It enables deterministic, auditable, and policy-governed agentic workflows with full observability.

The six new modules form the AI Execution Fabric:

ModuleRoleSprint
std/trace (schema)Canonical OTLP span hierarchy + attribute contractSprint 2
std/graphVirtual entity context assembly (3-tier)Sprint 3
std/flow (agentstep)AgentStep interface, Saga runner, HITL queueSprints 4 + 10
std/decision6-level governed execution veto ladderSprints 5, 6, 9
std/learnPluggable ML inference providerSprint 8
std/evalTrajectory replay and quality scoringSprint 11

Module Dependency Graph

graph TD
    T[std/trace schema] --> G[std/graph]
    T --> D[std/decision]
    T --> F[std/flow / agentstep]
    G --> D
    D --> L[std/learn]
    D --> F
    F --> E[std/eval]
    L --> E

Zero-CGO Constraint

All v2.0 core modules are compiled with CGO_ENABLED=0. No cgo dependencies are permitted in the pranor OSS repo. All heavy ML dependencies (PyTorch, TabPFN) run via:

  • Pure-Go WASM using wazero (no system calls required)
  • gRPC sidecar binaries over Unix domain sockets or TCP IPC

OSS / EE Build-Tag Convention

TagFile suffixBehavior
//go:build !enterprise_oss.goOSS implementation (stubs, in-memory)
//go:build enterprise_ee.goEnterprise implementation
(no tag)sharedInterfaces and types used by both

EE source lives in the pranor-ee repository under src/Pranor<Module>/.

v2.0-dev Branch Strategy

  • All v2.0 features are developed on the v2.0-dev branch of pranor
  • v1.0 development is frozen on main
  • v2.0-dev merges into main only after v1.0.0 is officially tagged
  • EE stubs in pranor-ee also track v2.0-dev

Sprint Completion Status

SprintIDFeatureStatus
1V2.89.0CI/CD Build Invariants✅ Complete
2V2.89.4Trace OTLP Span Schema✅ Complete
3V2.89.1Pranor Graph Module✅ Complete
4V2.89.3Flow AgentStep & Saga✅ Complete
5V2.89.2Graph Fault Tolerance✅ Complete
6V2.89.5Decision Engine (6-level)✅ Complete
7V2.89.6Decision Fault Tolerance✅ Complete
8V2.90.1Learn Provider Architecture✅ Complete
9V2.90.3Decision Simulation Engine✅ Complete
10V2.90.4HITL Approval Queue✅ Complete
11V2.90.2Pranor Eval Framework✅ Complete