The OSSA Open Roadmap: Community & Engineering

OSSA is moving from an experimental specification to a production-validated standard. This roadmap tracks our engineering progress across the specification, SDKs, and infrastructure substrates.

Current Specification: v0.3.6 (Alpha) Implementation Status: TS (Stable), Python (Active), Go (Planned)

Breakthrough: v0.3.6 Release (Jan 2026)

This week marks the transition to v0.3.6, which addresses the two biggest blockers for enterprise agent adoption: Cost and Identity.

1. Token Efficiency Framework (TEF)

We have implemented a multi-layered efficiency strategy designed to reduce the marginal cost of agent reasoning.

# v0.3.6 Efficiency Manifest
spec:
  efficiency:
    tier: economy
    promptCaching: true          # Support for Anthropic/OpenAI caching
    contextPruning:
      strategy: semantic-delta   # Only send what changed
      maxWindow: 4096
    budgets:
      hardLimitUsd: 0.05         # Declarative cost cap

Engineering Impact: In internal benchmarks on codebase-resident tasks, TEF combined with Knowledge Graph Indexing resulted in a 70-90% reduction in token consumption per task.

2. Functional Python SDK ([object Object])

Python is the lingua franca of AI. As of v0.3.6, our Python SDK is feature-complete with the TypeScript implementation.

from ossa import Agent, EfficiencyTier

# Load a declarative manifest
agent = Agent.from_yaml("security-scanner.ossa.yaml")

# Run with enforced efficiency tiers
result = agent.execute(
    input={"repo_url": "..." },
    tier=EfficiencyTier.ECONOMY
)
print(f"Task completed with exit code: {result.exit_code}")

3. Composite Identity & Attribution

We've moved beyond simple API keys to a Merged Identity Model, enabling full attribution across the agentic supply chain.

Identity Merging: Intersects permissions between the Agent Service Account and the Requesting User.
Audit Trails: Non-repudiable logs that satisfy SOC2 Type II and HIPAA audit requirements.

The Path to v0.4.0 (Beta - Q3 2026)

The next major milestone is the v0.4.0 Beta, focusing on externalizing policy and observability.

Q2 2026: The "Integration" Quarter

Go SDK Implementation: Porting core validation and execution logic to Go for high-concurrency substrates.
Agent-to-Agent (A2A) Protocol V2: Finalizing the JSON-RPC 2.0 handshake for context-preserving handoffs between frameworks (e.g., OSSA-Bridge for LangChain to OSSA-Bridge for CrewAI).
Formal Benchmarking: Publishing the first OSSA-Bench metrics comparing declarative vs. imperative agent performance.

Q3 2026: The "Governance" Quarter (v0.4.0)

Declarative Policy (Cedar/OPA): Moving from prompt-based instructions to externalized Policy-as-Code.
OTel Semantic Conventions: Standardizing how agent reasoning traces (thoughts, tool calls, corrections) are exported to OpenTelemetry collectors.
Security Guardrails: Built-in support for NeMo Guardrails and OWASP LLM Top 10 scanning.

Technical Success Criteria for v1.0 (Q4 2026)

Before freezing the API for v1.0, the specification must achieve:

Schema Stability: No breaking changes to the Core Kind (Agent, Flow, Tool) for 90 consecutive days.
Infrastructure Neutrality: Validated deployments on Kubernetes, AWS Lambda, and Vercel Edge.
Formal Verification: 100% of reference agents must pass the OSSA Conformance Suite.
Multi-SDK Parity: Identical behavior across TS, Python, and Go implementations.

Research & Standards Alignment

OSSA is designed to complement, not compete with, existing standards:

Model Context Protocol (MCP): OSSA uses MCP for tool-layer communication.
OpenTelemetry: OSSA extends OTel for agentic reasoning traces.
NIST AI RMF: OSSA Policies map directly to NIST risk management controls.

How to Contribute

We are an engineering-first community.

Specifications: Review the v0.3.6 RFCs on GitLab.
SDKs: Contribute to the Python or Go implementations.
Governance: Join the Technical Steering Committee (TSC) calls on the first Friday of every month.

Standardize once. Deploy everywhere.

Last Updated: January 29, 2026 Current Revision: v0.3.6-ENGINEERING