issus with OMNIROUTE

[salahkafa5-hub]

I tried to set up OmniRoute with Claude Code and test different models, but I encountered several issues:

• Some auto-* profiles selected OpenCode free-tier models and returned 403 errors because they can only be used from OpenCode.
• The glm52 profile pointed to an ambiguous Claude model instead of GLM.
• The zc-glm-5-2 profile generated 502 Stream ended before producing a non-ping SSE event.
• The actual ZCode model ID is zc/glm-5.2, but calling it directly returned 502: spawn zcode ENOENT, meaning the required zcode executable was not found.
• dva/glm-5-2 returned 500: DEVIN_AGENTIC_HOME must be an absolute path inside the bridge sandbox.
• cfp/zai-org/glm-5.2 returned a Cloudflare Playground browser/Playwright error because the required browser executable/session was unavailable.
• aug/glm-5.2 also returned 502 Stream ended before producing a non-ping SSE event.
• tllm/CLAUDE_4_6_SONNET returned 403 because the upstream provider was blocked by Vercel for the server egress IP.
• cxa/gpt-5.6-luna returned 503 because the Codex app-server transport was not configured.
• ddgw/claude-haiku-4-5 returned 418 because DuckDuckGo rejected the anonymous session due to an anti-abuse/rate-limit challenge.
• unc/qwen3.6:27b reported that the model does not exist or that access is unavailable.

The OmniRoute API itself works correctly: authentication works and /v1/models returns the model catalog. The main issue seems to be that several provider integrations require external CLIs, browser sessions, special environments, or provider-specific configurations that are not available or not correctly handled on Windows.

It would be helpful if the setup command could verify provider prerequisites and clearly indicate which models are actually usable with Claude Code before generating profiles.

[diegosouzapw]

Hi @salahkafa5-hub, thanks for the systematic list; it maps cleanly onto three different causes, and your closing suggestion is the right fix. I have opened #14363 for it.

1. The setup wrote profiles for providers your machine cannot run. setup-claude creates one profile per model in the live catalog (bin/cli/commands/setup-claude.mjs:78-131), and the catalog includes every no-auth provider, including the ones that only work through something outside OmniRoute:

• zc/ needs the ZCode app-server binary (spawn zcode ENOENT is exactly that), aug/ needs the Auggie CLI logged in, dva/ needs the Devin bridge sandbox, cxa/ needs the Codex app-server transport (docs/guides/CODEX-APP-SERVER-PROVIDER.md). The registry marks all four as local-CLI providers; the setup command does not read that flag yet, which is what the issue fixes.
• cfp/ drives Cloudflare's playground through a Playwright browser, and ddgw/ is an anonymous DuckDuckGo session that their anti-abuse challenge can reject (the 418).
• unc/qwen3.6:27b: UncloseAI's built-in ids go stale; use "Import from /models" on that provider to get the current list.

Until the gate ships, the clean way to get a usable set is: Settings -> Security -> Blocked Providers, block the prefixes above, then re-run omniroute setup-claude (or run it with --only glm,kmc,... for the providers you actually use).

2. OpenCode free tier (403 ... only be used from within OpenCode). Upstream contract change on their side; the fix (#14013) is merged on release/v3.8.51 and not in the published 3.8.50 yet. The auto/* profiles that picked OpenCode will recover on their own once you are on a build that carries it.

3. tllm/ was removed from the catalog at that provider's request (docs/reference/REMOVED_PROVIDERS.md); it exists in 3.8.50 and is gone from the next release, so the 403 there is not something to fix.

One thing I could not reproduce from the description: glm52 "pointed to an ambiguous Claude model instead of GLM". That profile is generated from the model id glm/glm-5.2, so its settings.json should read "model": "glm/glm-5.2". Could you paste ~/.claude/profiles/glm52/settings.json and what Claude Code showed when you launched it? If the id in the file is different, that is a separate bug and I want it.

[diegosouzapw]

@salahkafa5-hub -- still happening on the latest version (3.8.52)? Otherwise I'll close this in 7 days for inactivity.

[SHAJON-404]

@diegosouzapw still same problem