/ agents / omp

omp

Point the omp coding harness at base/mnt.ai with env vars.

omp is a multi-model coding harness that picks a model per task. It reads standard OpenAI-protocol env vars, so routing it through base/mnt.ai is one export.

Env vars

~/.zshrc

export OPENAI_BASE_URL=https://api.basemnt.ai/v1
export OPENAI_API_KEY=sk-your-key

If you route omp’s Anthropic-protocol tasks through base/mnt.ai as well:

~/.zshrc

export ANTHROPIC_BASE_URL=https://api.basemnt.ai
export ANTHROPIC_AUTH_TOKEN=sk-your-key

Bash function

~/.zshrc

omp() {
  OPENAI_BASE_URL=https://api.basemnt.ai/v1 \
  OPENAI_API_KEY="$SK_KEY" \
  command omp "$@"
}

Model selection

omp selects models per task. Give it catalog IDs — z-ai/glm-5.2 for heavy refactors, deepseek/deepseek-v4-flash-0731 for cheap mechanical passes — through omp’s model configuration or -m flag, depending on version.

Verify

Launch omp on a repo task and watch your usage dashboard increment per request. A 401 means the key variable isn’t exported in the harness’s environment; note that spawned subagents inherit the parent’s env, so one export covers them all.