Integrations
Use Venkai with your existing stack.
Every row carries a status and the evidence behind it. "Compatible by architecture" means no official integration exists and nothing has been tested.
Where it plugs in
Next to what already runs
- 01ApplicationYour product, unchanged.
- 02Agents and stepsYour orchestrator, models, memory and RAG stay where they are.
- 03Venkai — REST · SDK · MCPCalled by the agent, next to those systems rather than underneath them: read before the prompt, write after the decision.
- 04Typed work stateFact, decision, constraint, preference, event, relationship — versioned, isolated per organization.
- 05Work continuesThe next agent, session or model resumes from that state.
Legend
Four statuses, never mixed
Supported
Official interface, documented, covered by tests.
Tested
Actually exercised with Venkai, with a named piece of evidence.
Compatible by architecture
Nothing in the architecture prevents it. No official integration, no test.
Planned
On a dated roadmap. No integration is on it today.
Status
Venkai interfaces
| System | Status | What is actually verified |
|---|---|---|
| REST API | Supported | Documented (/api/memory, /api/context/{project}/relevant). Covered by tests/test_api.py. |
| Python SDK | Supported | remember / recall / decisions / timeline. Covered by tests/test_sdk.py and the handoff demo (19/19). |
| MCP (stdio) | Tested | Beta. Covered by tests/test_mcp.py; used daily by Claude Code inside the founder's own agent system. |
Status
Memory
| System | Status | What is actually verified |
|---|---|---|
| Mem0 | Compatible by architecture | No adapter, no test. An agent can call both in the same step; nothing in Venkai assumes or excludes this system. |
| Zep | Compatible by architecture | No adapter, no test. An agent can call both in the same step; nothing in Venkai assumes or excludes this system. |
| Letta | Compatible by architecture | No adapter, no test. An agent can call both in the same step; nothing in Venkai assumes or excludes this system. |
| Custom memory systems | Compatible by architecture | Venkai is two HTTP calls and depends on no other store. Not tested with a third-party system. |
Status
Orchestration
| System | Status | What is actually verified |
|---|---|---|
| LangGraph | Compatible by architecture | Documented pattern: a node before the model loads context into state, a node after it writes conclusions. No package, no test. |
| CrewAI | Compatible by architecture | Documented pattern: a task callback, or a custom tool the agent calls. No package, no test. |
| AutoGen | Compatible by architecture | No documented pattern, no test. Same principle: read before the prompt, write after the decision. |
| Custom orchestrators | Tested | The documented loop (load before the prompt, record after the decision) runs in the A → B → C handoff demo: 19/19 checks, real API server, Python SDK. |
Status
Models
| System | Status | What is actually verified |
|---|---|---|
| Claude | Tested | Through Claude Code and the MCP server, in daily in-house use. Not through the Anthropic API directly. |
| GPT | Compatible by architecture | Venkai calls no model and returns JSON. No agent running this model has been tested against Venkai. |
| Gemini | Compatible by architecture | Venkai calls no model and returns JSON. No agent running this model has been tested against Venkai. |
| Local models | Compatible by architecture | Venkai calls no model and returns JSON. No agent running this model has been tested against Venkai. |
Priorities
What decides the next integration
No integration gets built on intuition. The order will follow two signals: the stacks described in pilot requests, and measured interest on this page and in the FAQ. If your stack is missing or only compatible, say so in the pilot request — that is exactly the signal.
Tell us which stack you run.
Memory, orchestrator, models, and where context gets lost. We answer with a pilot scope, or with the reason Venkai isn't the right piece.
Describe your stack