Repository navigation
[Question] Policy on third-party open-source apps using Zen free-tier models #49144
Description
Feature hasn't been suggested before.
- I have verified this feature I'm about to request hasn't been suggested before.
Describe the enhancement you want to request
Hi opencode team — we're building a free IT assistant tool: a non-commercial, open-source (MIT-planned) Windows maintenance
assistant for office workers with no IT skills (adware cleanup, startup optimization, troubleshooting guidance).
TL;DR: we'd like to offer Zen free-tier models as our zero-config default, and to vendor your @opencode-ai/llm engine package
(MIT) as our LLM request layer. Before doing either, we want your explicit blessing.
Questions:
- Is third-party open-source use of the Zen free tier permitted under your ToS, with or without restrictions (rate limits,
attribution, specific User-Agent)? - If yes, is there a sanctioned way to identify our app (e.g. a dedicated header or UA string you would whitelist)?
- Are you comfortable with us vendoring @opencode-ai/llm in this way? Any requests beyond the MIT license terms (attribution
wording, a note that we're unaffiliated, etc.)? - If the free tier is off-limits for third parties, would a paid Zen API key be the recommended path for an app like ours?
We're happy to comply with whatever policy you set — attribution, linking to opencode, rate limiting, or staying out entirely.
Thanks for the great work on opencode.
Background & context (click to expand)
Who our users are. Office workers with no IT skills at all — when their PC slows down or gets hijacked by adware, they usually
just live with it, or fall for overpriced "PC cleaner" scams. We want to give them a trustworthy free tool that actually fixes
these problems, explained in plain language they can understand.
How we use LLMs. (a) intent classification and explaining diagnostic results in plain language; (b) reasoning about
troubleshooting approaches for problems our local knowledge base can't cover; (c) occasionally drafting short
analysis/remediation scripts (e.g. Python) to locate or fix an issue. Safety note: all consequential decisions are made by a
local deterministic rule engine, and every script or action goes through a mandatory user-confirmation gate with audit logging
before execution — the LLM never executes anything by itself.
What we observed. The free tier currently requires the x-opencode-session header (400 MissingSessionID, "OpenCode's free tier can
only be used in OpenCode"). We have NOT worked around this and won't: we would never ship the session-layer headers or present
ourselves as the opencode client — requests from our app would be honestly identified however you prefer.
Vendoring. @opencode-ai/llm is MIT licensed — thank you. If we vendor it, your copyright notice stays intact and our README will
carry clear attribution plus a note that we're unaffiliated.
Activity
edge513 commentedon Sep 16, 2026
Adding a concrete user case: I use Hermes Agent (NousResearch/hermes-agent) and would like to access Union Alpha through OpenCode Zen from Hermes. I am asking as a user, not on behalf of the Hermes maintainers.
The Zen documentation lists
union-alphaathttps://opencode.ai/zen/v1/messages. A minimal unauthenticated request using that Messages API, honestly identified as Hermes and without anx-opencode-sessionheader, returned HTTP 400:{"type":"error","error":{"type":"MissingSessionID","message":"Error from provider (Console): OpenCode's free tier can only be used in OpenCode"}}Could the team clarify:
I am looking for an officially supported integration, not a workaround for client restrictions or free-tier limits. Thanks for clarifying!
cosmichut commentedon Oct 11, 2026
Follow-up in the spirit of transparency (a one-line objection is enough — no detailed response needed):
Since we haven't heard back, here is the approach we now plan to take:
opencode serve). All requests to Zen would be issued by the genuine opencode process itself, via its native request path — nothing trimmed from request building, credentials, tool surface, or User-Agent.bash/edit/write: deny). We only consume the model's text output; every consequential action in our app goes through our own confirmation + audit layer.If this form of bundling is against your policy, just say so here and we will drop the Zen path entirely and stay BYOK-only. Thanks again for opencode.
Tarquinen commentedon Oct 20, 2026
Thanks for reaching out! OpenCode's free models are meant to be used only from within OpenCode's own clients, so direct API calls and third-party apps are rejected on purpose. For API access from other tools, please use the paid models on OpenCode Zen or Go: https://opencode.ai/v2/docs/console/go. Closing this issue.
— from 𝕺𝖕𝖊𝖓𝕮𝖔𝖉𝖊