Skip to main content
Back to Research
UNDEFINED
Specification

Introducing OSSA: The OpenAPI for AI Agents

OSSA is the formal specification standard for defining AI agent manifests. We analyze how OSSA addresses the 'Fragmentation Crisis' in the agentic ecosystem through declarative contracts and vendor-neutral protocols.

Thomas Scola··3 min read

Introducing OSSA: The OpenAPI for AI Agents

The AI agent ecosystem is facing a Fragmentation Crisis. Every framework—LangChain, CrewAI, AutoGen, and proprietary provider labs—has defined its own way of describing an agent. This coupling of intent to implementation is the exact problem that OpenAPI solved for REST APIs a decade ago.

OSSA (Open Standard for Software Agents) is the declarative contract for the agentic web.

What is OSSA?

OSSA is not a library or an execution engine. It is a Formal Specification governed by a community of engineers. It provides a shared language for defining the four pillars of an autonomous actor:

  1. Identity: URI-addressable, DID-backed existence.
  2. Capabilities: Abstract tool interfaces using the BAT (Bridge, Adapter, Tool) pattern.
  3. Governance: Declarative cost caps, RBAC scopes, and compliance mapping.
  4. Orchestration: Standardized signals for multi-agent handoffs.

The OSSA Manifest Anatomy

A production OSSA manifest (v0.3.6) separates governance from the prompt:

apiVersion: ossa/v0.4.0
kind: Agent
metadata:
  name: "fin-ops-auditor"
spec:
  # The Reasoning Core
  role: "Audit financial transactions for compliance"
  llm:
    provider: google
    model: gemini-1.5-pro
  
  # The Capability Layer
  capabilities:
    - protocol: mcp
      server: postgres-mcp-server
      tools: ["query_ledgers"]

  # The Governance Layer
  security:
    id: did:ossa:auditor-v1
    scopes: ["read:ledger", "write:audit-log"]
  efficiency:
    tier: economy # Forces prompt caching and context pruning

Why Standardization is Mandatory Now

The "Move Fast and Break Things" phase of AI agents is over. Enterprises require Stable Substrates.

1. Eliminating Vendor Lock-in

Without OSSA, your agentic logic is trapped in framework-specific code. With OSSA, you can define your agent once and switch from LangChain to CrewAI (or from OpenAI to Anthropic) with zero code changes. You simply point a new "Bridge" at the OSSA manifest.

2. Solving the "Vibe-Check" Validation Problem

OSSA manifests are validatable against JSON Schema. This enables CI/CD pipelines to catch logical errors (missing tools, incorrect model tiers, or budget violations) before a single token is spent.

3. Federated Accountability

As agents move from simple chat bots to "First-Class Citizens" that execute financial transactions and modify production code, we need a standard for Lineage. OSSA’s Composite Identity model ensures every action is attributed to both the agent and the human owner.

Alignment with Research & Industry

OSSA is designed to be the "glue" between existing research breakthroughs:

The Path Forward: v1.0 and Beyond

We are currently in the v0.3.x cycle (Alpha), with a focus on production-validated features. Our Engineering Roadmap targets a v1.0 stable release by Q4 2026.

Are you building for the long-term? Stop writing imperative agent loops and start defining declarative contracts.

Technical Resources:

StandardsInteroperabilityGovernanceOSSA
Back to All Research