AgentBox marketplace deployment (historical)

The removed AgentBox container surface, kept for anyone debugging an older build.

OpenHuman once shipped a container surface for the GMI Cloud AgentBox marketplace. It is gone from openhuman-core. A regression test, agentbox_run_and_jobs_paths_are_no_longer_public in crates/openhuman-rpc/src/server/auth_tests.rs, checks that /run and /jobs/{job_id} no longer exist.

OPENHUMAN_AGENTBOX_MODE has no reader left in crates/openhuman-core/src. The Dockerfile and .env.example still mention it, and those mentions are stale. If you need to run OpenHuman on AgentBox today, ask the team that owns the GMI Cloud deployment where that surface lives now.

The rest of this page describes how the old surface worked. Read it only if you are debugging an older build that still has it.

Container contract

With OPENHUMAN_AGENTBOX_MODE=1, the core HTTP server exposed three routes:

• POST /run accepts work and returns 202 { "job_id": "<uuid>" }. The body is { "payload": { "message": "<string>", "thread_id": "<optional string>" } }.

• GET /jobs/{job_id} returns { "status": "pending|running|completed|failed", "result": ..., "error": ... }.

• GET /health is a liveness check.

/run and /jobs/* were unauthenticated at the container boundary. AgentBox's edge handled auth before traffic reached the container.

Registering in the AgentBox console

The console has a four-step wizard.

1. Basic info. Name it OpenHuman and add a description.

2. Infrastructure. Pick the Docker image source, compute tier and region. Turn on the "GMI MaaS" toggle so the platform injects GMI_MAAS_BASE_URL and GMI_MAAS_API_KEY at runtime.

3. Env variables. Set these:

   • OPENHUMAN_AGENTBOX_MODE=1

   • OPENHUMAN_AGENTBOX_JOB_TIMEOUT_SECS (optional, default 600)

   • GMI_MODELS, the marketplace-approved model id, for example deepseek-ai/DeepSeek-V4-Pro

   • OPENHUMAN_WORKSPACE, a writable container path such as /home/openhuman/.openhuman

   • RUST_LOG=info (use debug for the first deploy)

4. Review and register. Confirm, then test from the console panel.

The platform API key is shown once, on the registration confirmation screen. Save it to your secrets manager right away. The console cannot show it again.

Pushing the image

Build from main with the existing Dockerfile and push to your registry:

docker build -t <registry>/openhuman-core:<tag> .
docker push <registry>/openhuman-core:<tag>

The first deploy takes 10 to 25 minutes to reach running. Later deploys are faster.

Long-running requests

AgentBox treats requests over 2 minutes as long-running, so OpenHuman used polling. The agent ran inside a worker task, capped by OPENHUMAN_AGENTBOX_JOB_TIMEOUT_SECS (10 minutes by default). There was no streaming.

A polling client should:

1. Send POST /run and keep the job_id.

2. Call GET /jobs/{job_id} every 1 to 3 seconds.

3. Stop when status is completed or failed.

Finished jobs were kept for one hour, then garbage-collected. A long pause between polling and reading can return 404.

Local smoke test

OPENHUMAN_AGENTBOX_MODE=1 \
GMI_MAAS_BASE_URL=https://api.gmi-serving.com \
GMI_MAAS_API_KEY=sk-... \
GMI_MODELS=deepseek-ai/DeepSeek-V4-Pro \
./target/debug/openhuman-core serve &

curl -X POST http://127.0.0.1:7788/run \
  -H 'content-type: application/json' \
  -d '{"payload":{"message":"hello"}}'

# Then poll the returned job_id:
curl http://127.0.0.1:7788/jobs/<job_id>

Troubleshooting

• 404 job not found after a successful submit: the one-hour retention window passed, or the container restarted. The job store was in memory only.

• status: "failed" with error: "agentbox: agent runtime bridge not wired": you are on a build from before the runtime bridge landed. Rebuild from a current main.

• status: "failed" with error: "job timeout after Ns": the agent ran past OPENHUMAN_AGENTBOX_JOB_TIMEOUT_SECS. Raise it on the next deploy.

• [agentbox::gmi] not registering GMI MaaS provider: missing/blank: GMI_MAAS_API_KEY: the platform did not inject the key. Check the MaaS toggle in step 2 of the wizard.

• [agentbox::gmi] current-thread runtime detected, skipping provider registration: the core booted in a single-threaded tokio runtime. Use the standard serve subcommand, which starts a multi-thread runtime.