Comparison¶
How AgentSpec compares to similar projects.
vs. gitagent (Lyzr AI)¶
| Feature | AgentSpec | gitagent |
|---|---|---|
| Manifest format | YAML, multi-format | YAML |
| Resolver | ✓ auto-negotiates env | ✗ |
| Inheritance | ✓ enforced merge strategies | partial (URL-based extends) |
| Trust enforcement | ✓ hardcoded restrict | ✗ |
| Signing/profiles | ✓ Ed25519 portfolios | ✗ |
| Registry | ✓ via Noether | proprietary |
| Owner | community (EUPL-1.2) | Lyzr AI (single company) |
gitagent has a clean format but no resolver, no profile system, and is tied to one vendor.
vs. Agent Format (Snap Inc.)¶
| Feature | AgentSpec | Agent Format |
|---|---|---|
| Spec quality | ✓ comprehensive | ✓ clean |
| Resolver | ✓ | ✗ |
| Inheritance | ✓ | ✗ |
| Tooling | ✓ CLI + SDK | spec-only |
| Profiles | ✓ | ✗ |
| Active | ✓ | abandoned-ish |
Agent Format had a great spec but no working tooling. AgentSpec borrows the cleanest ideas and adds the missing pieces.
vs. OSSA¶
| Feature | AgentSpec | OSSA |
|---|---|---|
| Lightweight | ✓ pip install | ✗ Drupal-based |
| Cryptographic verification | ✓ Ed25519 | ✓ chains |
| Discovery | semantic search | DNS-native |
| Complexity | medium | high |
| Audience | developers | enterprises |
OSSA is over-engineered for most use cases. AgentSpec aims for the 80% with less ceremony.
vs. LangGraph / CrewAI / AutoGen¶
These are orchestrators, not standards:
| Feature | AgentSpec | LangGraph/CrewAI/AutoGen |
|---|---|---|
| Manifest format | ✓ universal | each has its own |
| Cross-orchestrator portability | ✓ | ✗ |
| Resolver | ✓ | partial |
| Profiles | ✓ | varies |
AgentSpec is complementary. You can use LangGraph/CrewAI/AutoGen as the orchestrator and AgentSpec as the agent definition format. The .agent file becomes portable across orchestrators.
vs. MCP (Model Context Protocol)¶
MCP is about tool/resource exposure to agents. AgentSpec is about agent definition.
| Layer | Standard |
|---|---|
| Tools/resources for agents | MCP |
| Agent definition | AgentSpec |
| Orchestration | LangGraph, CrewAI, caloron-noether, ... |
| Composition | Noether |
They stack:
caloron-noether (orchestrator)
↓
agentspec (defines the agents)
↓
mcp servers (provide tools to the agents)
↓
noether (composes verified stages)
Use them together.
vs. Docker / OCI¶
The analogy is direct:
| Docker / OCI | AgentSpec |
|---|---|
Dockerfile |
.agent file |
docker build |
agentspec resolve |
docker run |
agentspec run |
docker pull/push |
agentspec pull/push |
| Docker Hub | registry.agentspec.dev (via noether-cloud) |
| Image hash | content hash (ag1:xxx) |
| Image layers | inheritance chain |
But agents are not containers — they're declarations resolved at runtime. The closest analogy is Helm charts for AI agents: declarative, parameterized, registry-distributable.