BUG OrcaRouter

[hana189]

orcarouter/kimi/kimi-k3] [404]: Invalid URL (POST /v1) (reset after 1m 24s)

Hy3 , deepseek flash and pro too

[diegosouzapw]

Hey @hana189! This looks like the model ID string, not an OmniRoute bug on the routing side -- OrcaRouter's own distributor is strict about namespaced model IDs (their code comment notes even a bare auto returns an error if not prefixed correctly).

OrcaRouter's pinned catalog in OmniRoute only exposes these IDs today: orcarouter/auto, openai/gpt-5.5, google/gemini-3.5-flash, anthropic/claude-opus-4.8, grok/grok-4.3, deepseek/deepseek-v4-pro, minimax/minimax-m2.7, qwen/qwen3.7-max. There's no kimi-k3 or a "deepseek flash" in that list -- could you paste the exact model ID string you configured in the combo/connection? If OrcaRouter added new models to their catalog that we haven't pinned yet (the provider supports pass-through), I can get them added; if it's a typo/wrong prefix, I can point you to the right one.

[hana189]

[404]: Invalid URL (POST /v1) the mode im use auto import canot use

[diegosouzapw]

@hana189 -- thanks for the screenshot. That [404]: Invalid URL (POST /v1) with the "auto import" mode is a base URL problem, not a model-ID one, so my first guess about the model string was the wrong thread to pull. Apologies.

What's happening: "auto import" is trying to POST to /v1 directly -- there's no /v1/chat/completions path appended. That means the base URL the connection (or the import) was given is missing or malformed, so the executor builds a path that OrcaRouter's gateway rejects as an invalid route.

Two things to check on the OrcaRouter connection in Dashboard -> Providers:

1. Base URL field -- it should be the gateway root with no trailing path. For OrcaRouter that's normally just their API host (no /v1 appended -- OmniRoute adds /v1/chat/completions itself). If it has a stray /v1 or is empty, you get exactly this POST /v1 shape.
2. The model ID -- once the base URL is correct, the model still needs to be a namespaced one OrcaRouter's distributor recognizes. auto alone returns 503 on their side unless prefixed, so use the pinned orcarouter/auto (or a specific one like orcarouter/openai/gpt-5.5).

Could you paste the Base URL value exactly as it sits in the connection config? If it's the host with /v1 already on it, that's the bug -- strip the /v1 and the import should resolve.

[hana189]

Oke work add base url manually thanks @diegosouzapw . One more question how to edit slug Openai-compatible-chat-c52ubs59-3ewc-44db-b49a-5abwj7a8a142d on ui web? Because auto discovery is show too length in agent. Add manually provider on ui get slug like that.

[diegosouzapw]

Glad the manual base URL fixed the OrcaRouter request, @hana189. There is no dashboard control today for renaming that generated openai-compatible-chat-... model slug. The upstream model ID and the client-facing alias are currently tied too closely in this discovery path.

I opened #9068 for an editable, stable alias that keeps the original upstream model ID on the wire. That should solve the readability problem without breaking routing when models are rediscovered.

[hana189]

Thanks, i hope that done on next release. I can't wait for it.
And @diegosouzapw how to fix this? Im now istalled omniroute on my coolify server. But cline connect for free account need cli? Codex or grok no installed cli on my server its fine not get warning like that. Is that warning okey? Sorry bad english 😅

[diegosouzapw]

Quick answer, @hana189: the warning is expected and harmless on a headless server. You can ignore it.

What it is: the CLI Code tab (src/app/(dashboard)/dashboard/cli-code/) detects local binaries -- the Cline CLI/VS Code extension on the operator's machine -- so it can auto-write ~/.cline/data/globalState.json for you. On a server with no Cline IDE/CLI installed, it always reads "not detected", same as Codex and Grok (src/shared/services/cliRuntime.ts:95-105 shows cline: requiresBinary: true next to codex/grok-build, all the same shape).

What it is not: it does not mean you cannot use Cline through OmniRoute on Coolify. Cline works server-side via two paths that need no local binary:

• OAuth: src/lib/oauth/providers/cline.ts (authorization_code flow against api.cline.bot, refresh also server-side in open-sse/services/tokenRefresh/providers/cline.ts:18).
• API key: paste the key directly in the connection.

Add the connection under Dashboard -> Providers -> Cline, choose OAuth or API key, and route normally. The CLI Code banner stays yellow but everything else works. See docs/guides/REMOTE-MODE.md for the full headless setup.

[hana189]

Oh okey just ignored, and @diegosouzapw i want add qoder pat and get error

Its qoder need install cli? And how to install on coolify inside omni?
And kr/claude-sonnet-4.5 from kiro its not show from auto discovery models. Eye icon im click test Enable disable not work too on ui web.

[diegosouzapw]

Hey @hana189! Two open threads in the same Discussion -- let me address the latest pair:

1. Qoder PAT install on Coolify -- Qoder is a CLI-required provider, same family as Codex and Grok. You need to install the Qoder CLI inside the OmniRoute container (or mount the binary + persistent ~/.qoder from the host), set CLI_QODER_BIN to the path, and then add the Qoder PAT connection in the dashboard. The "Local CLI runtime not installed" warning means the executable was not located. Detailed steps in docs/guides/REMOTE-MODE.md. The recent [502] symptom you saw on #9267 is tracked in #9319.

2. kr/claude-sonnet-4.5 not showing on auto-discovery + Eye toggle not working -- could you share the OmniRoute version and which discovery source you are using (Dashboard -> Providers -> Kiro)? kr/claude-sonnet-4.5 should appear on the OAuth-based Kiro provider. If it does not, your install may be running an older catalog -- pull the latest image and re-import models. The Eye (enable/disable) toggle has a known UI bug in the custom-provider flows for very long slugs; we have a fix staged. If you can paste the version + a screenshot of the model list, I will attach it to the existing tracking issue.