Gateway shows "Model catalog unavailable" for large model catalogs — please upgrade to OMP RPC v2 negotiation

[TTcheng]

PinkCollab 3.0.8 + OMP 18.10.0, ~600 models in the catalog (openrouter 568 + minimax-code-cn 10)
→ the Android client reliably shows "Model catalog unavailable" when switching models or
thinking levels.

Not a bug in the PinkCollab UI, not a bug in OMP — Gateway currently spawns OMP over RPC v1 and
hits OMP's hard cap of MAX_RPC_FRAME_BYTES = 1 MiB. OMP rejects the get_available_models
response itself, and the failure surfaces in the UI as that string.

OMP already ships an answer (chunked framing on v2), it just isn't being asked for:

· can1357/oh-my-pi#7877 (OMP upstream)
— get_available_models hard-fails on v1 once the catalog exceeds 1 MiB.
Maintainer tagged it wontfix; the only sanctioned fix is at the protocol layer.
Same catalog succeeds in 5 chunks over v2 (~1.07 MiB total).
· getpaseo/paseo#2966 (sibling client that hit the same wall)
— Title says it directly: "Paseo never negotiates OMP/Pi RPC protocol v2."
Fix: consumer sends
{"type":"negotiate_protocol","protocolVersion":2}
on startup, then reassembles rpc_chunk sequences by byteLength + chunk index.
· OMP docs/rpc.md documents the contract: v2-capable clients "should immediately" send
negotiate_protocol.

=======================

Local fallback: dropped openrouter from disabledProviders, catalog now 10 models and works.
Temporary — the same wall reappears the moment the catalog grows again.

=======================

Two questions for @3xian:

1. Is Gateway upgrade to v2 negotiation on the roadmap?
2. If not, I'd like to fork gateway-linux-x64 myself to add negotiate_protocol. Could
   you share a forward-port note (target branch, manifest entries, release flow for the
   prebuilt artifact) so a PR can land cleanly?

Note: this post was drafted in Chinese and translated

[3xian]

Thanks for the detailed investigation and the upstream references!

You're right, this should be fixed in the Gateway. I'll add RPC v2 negotiation and chunked response handling as soon as possible.

No need to fork the Gateway for this. I'm working on the fix now and will update this discussion once it's available.

Thanks again for reporting this!

[TTcheng]

Thanks for confirming. Looking forward to the fix and happy to test it once it ships.

[3xian]

Fixed on main and released in v3.1.0.

[TTcheng]

I’ve already tried it. The problem has been solved and everything is working fine. Thanks again.