Skip to content

Update to https://github.com/serenorg/seren/issues/135 Still getting error #139

@terraleiloa

Description

@terraleiloa

Comment 1 — Follow-up: on_demand still rejected (after Seren update)
When to post: After you heard Seren was updated to support on_demand.

We tried deploying again with mode: "on_demand" after the update. The deploy API still returns:

Request: POST https://api.serendb.com/publishers/seren-cloud/deploy with "mode": "on_demand"
Response: 400 Bad Request — "Invalid mode 'on_demand'. Must be 'always_on' or 'cron'."
So from our side it still looks like only always_on and cron are accepted. Can you confirm whether:

The new mode is live on api.serendb.com (or if we should use a different endpoint)?
The exact mode value the API expects (e.g. on_demand vs on-demand vs something else)?
Once that’s clear we can redeploy as on_demand and use POST .../run for one-shot invokes. Until then we’re blocked on always_on (no invoke URL, run returns 500).

Comment 2 — Short base URL for others hitting this
When to post: Optional; helpful for anyone else trying the APIs.

For anyone calling the control-plane today, the short base URL works:

Base: https://api.serendb.com/publishers/seren-cloud
GET agent: GET .../agents/{id}
POST run: POST .../agents/{id}/run (with JSON body; returns 500 for always_on agents)
Deploy: POST .../deploy
So the issue isn’t the base path — it’s that for always_on we have no way to actually invoke the skill (no invoke URL in the response, and run tries to create a Job and fails).

Comment 3 — What would resolve this for us
When to post: Optional; summarizes what would unblock you.

What would resolve this for us (in order of preference):

Invoke URL in deploy/agent response — Deploy and/or GET .../agents/{id} return a field (e.g. endpoint, invoke_url, or url) with the base URL to POST { "state", "input" } to for always_on agents.
Run proxies to always_on — POST .../agents/{id}/run with a body is proxied to the running container and returns the skill response (e.g. { state, prompt, outputs? }) instead of creating a K8s Job.
Supported on_demand mode — If the API accepts an on_demand (or equivalent) mode, we can deploy that way and use run for one-shot invokes; we’d just need confirmation that it’s live and the exact mode string.
Any one of these would unblock us. Thanks for looking into it.

Comment 4 — One-line “still seeing 400” follow-up
When to post: Quick follow-up if someone says on_demand is deployed.

Tried again just now: POST .../deploy with "mode": "on_demand" still returns 400 — "Invalid mode 'on_demand'. Must be 'always_on' or 'cron'." So we’re still unable to deploy as on_demand from here.

QUESTION:
ARE AGENT SKILLS REQUIRED TO BE DEPLOYED AS PUBLISHERS?

Metadata

Metadata

Labels

bugSomething isn't working

Type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions