OMO Slim with local models only

[acaleechurn]

Any one using OMO Slim with local models and have you been able to configure the maximum number of agents spawned by the Orchestrator? I've been struggling to prevent it from launching multiple instances of the main coder (overloading the GPU) instead of queuing the tasks and ended up creating an MCP server to act as a concurrency gate. Anyone with a better idea? I use Qwen 3.6 35BA3 as the Orchestrator, Qwen 3.6 27B as the main coder and Gemma 4 26B as the designer across 3 GPUs. It works but I'm not sure if it's the right call.

[mhenke]

There's no maxConcurrency setting. To control parallelism on local models, tune the orchestrator prompt to serialize delegation, or gate concurrency at your MCP layer. See also the related tip in #594.

[acaleechurn]

@mhenke Thank you for the reply. I saw related tip earlier and initially tried with an orchestrator_append.md file but it directly contradicted OMO's own built-in “spawn multiple fixers” guidance, so it was ignored in practice. I then tried the same file, rewritten to explicitly override that built-in guidance with a mandatory checklist but it was still text instructions, not enforcement. So the MCP concurrency gate was the only working option. I have attached it in case someone faces the same issue.

concurrency-gate.mjs.txt

I just include it in the opencode.json configuration:

"concurrency-gate": {
"type": "local",
"command": [
"sh",
"-c",
"exec /home/acaleeechurn/.bun/bin/bun /home/acaleechurn/.config/opencode/mcp/concurrency-gate.mjs"
]