Is there a roadmap published somewhere that I missed?

[dbrewster]

Sorry if there is. I'm looking to see when the remaining items that are 3.5 around command line execution, swarm deployments, etc.. are planned to be finished. Currently lots of those items are just stubs.

I'm happy to help out where needed. I have lots of experience writing agent frameworks.

[roman-rr]

You're right that lots of items are stubs — and your offer to help is generous.

We audited the full codebase to understand exactly what's implemented vs planned. Here's the current state as of v3.5.51:

Implemented (actually executes):

• Memory/HNSW vector search (memory_store/search/retrieve)
• Embeddings (embeddings_generate/compare)
• Terminal execute
• Session save/restore
• Browser automation (if Playwright connected)

Stubs (JSON state, no executor):

• Agent lifecycle (agent_spawn, status, pool) — Map entries only
• Task management (task_create, assign) — EventEmitter with no listener
• Swarm coordination — config storage, agentCount always 0
• All consensus types — same JSON handler regardless of selection
• Neural training — ignores data, random accuracy
• WASM agents — echoes input
• Workflows — state machine with no executor
• Coordination, Claims, DAA — registration stubs

The key architectural gap: AnthropicProvider (real LLM calls), task queue, and agent registry all exist — but nothing wires agent_spawn to LLM execution. That's the single biggest piece to connect.

No public roadmap exists that we could find. The closest is the maintainer's responses on issue #653 (claiming improvements) and #1383, but the stubs remain in the current release.

Full audit: Independent Audit