Agent Compatibility (full attestation pending)

The integration surface is available now

Agent Compatibility's capabilities ship today: recipes, candidate-write APIs, MCP tool exposure, the SDK helper, and observability logging are all reachable, and every agent write is review-gated. The full "live" status, backed by a four-runtime manual attestation, is still planned and stays gated on the manual attestation reaching PASS.

• Canonical surface, /docs/concepts/agent-compatibility
• Verification gate, manual four-runtime attestation (currently PENDING)
• Recipes are linked from the concept page and from the Recipes sidebar group.

What it is

• Recipes for ADK / LangGraph / OpenAI Agents SDK / OpenClaw / MCP clients.
• Per-agent identity templates + scope recipes.
• Observability hook examples for agent runs.
• Memory-lifecycle rules + review-before-save patterns.

What it is not

• Not a full Agent OS.
• Not a scheduler or orchestrator.
• Not a runtime or sandbox.
• Not autonomous execution.

Where Shelvia fits

Agent runtimes live in ADK / LangGraph / OpenAI Agents SDK / OpenClaw / MCP clients. Shelvia is the memory layer they call, before they act.

Recipes

These recipes document four agent-runtime integrations plus the cross-cutting candidate-write pattern. Each recipe is a docs-only page that uses the already-shipped REST, MCP, and SDK surface. No new Shelvia runtime is needed.

• OpenAI Agents SDK, /docs/recipes/openai-agents-sdk
• MCP client (Claude Desktop / Cursor / Cline / Continue), /docs/recipes/mcp-client
• LangGraph, /docs/recipes/langgraph
• Google ADK, /docs/recipes/google-adk
• Agent proposes memory candidate, /docs/recipes/agent-proposes-candidate