Demon Route

🌍 View in other languages

Demon Route (API-key gateway)

Demon Route provides an OpenAI-compatible, externally hosted model gateway. Its catalog contains open-weight and independently fine-tuned models; there are no OpenAI/Anthropic/ Google closed-model SKUs. See the official API docs, catalog, and pricing for current model IDs, costs, entitlement and privacy policy.

Configure a connection

1. Create a Demon Route account and an API key on demonroute.com. A confirmed deposit is required for paid models and to unlock its conditional complimentary lane. OmniRoute deliberately does not mark this provider as hasFree: the complimentary lane is not unconditional.
2. In Providers → API Key Providers → Demon Route, create a connection and paste the key. The built-in endpoint is https://api.demonroute.com/v1; do not enter the /chat/completions suffix as an OpenAI-compatible node URL.
3. In the connection's Available Models, choose Import from /models or enable Auto-Sync to obtain the models your key can access. The small bundled seed list is for offline fallback, not proof of current entitlement, model health or price.
4. Choose an exact, case-sensitive upstream channel ID, retaining its vendor prefix (for example dphn/Dolphin3.0-Llama3.1-8B). On the OmniRoute API, prefix it once with demonroute/, giving demonroute/dphn/Dolphin3.0-Llama3.1-8B. The extra slash is intentional: the upstream vendor/model string must not be truncated or rewritten.
5. Test both non-streaming and streaming chat via the dashboard before inserting it as a member of a fallback combo. Tool support, context lengths, modalities and pricing vary by model; do not assume every model supports tools or vision.

Request example

With a separate OmniRoute gateway API key, a local non-streaming smoke request is:

curl -sS http://localhost:20128/v1/chat/completions \
  -H "Authorization: Bearer ${OMNIROUTE_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{"model":"demonroute/dphn/Dolphin3.0-Llama3.1-8B","stream":false,"messages":[{"role":"user","content":"Say hello."}]}'

Use the same endpoint with "stream":true and curl -N to verify SSE. The key above authenticates the client to OmniRoute. Demon Route's own Bearer key belongs only in its stored provider connection; do not paste the provider key into source code, combo definitions, logs, or screenshots. Smoke requests are real upstream requests and may consume balance or daily quotas.

Endpoint support and operational boundaries

Surface	Behavior
Chat Completions	/v1/chat/completions, JSON and SSE streaming
Model discovery	/v1/models, requires the connected Bearer key; import the live per-key list
Embeddings	Not declared: no embedding-capable model is verified for this gateway, so it is not listed as an embedding provider
Responses, images, speech, video	Not advertised by this integration: no provider support is inferred merely because it resembles OpenAI
Timing	120-second request and first-byte cap in the built-in configuration; upstream advises clients to allow at least 120 seconds
Balance / quotas	Prepaid gateway; 402 indicates insufficient credit, 429 indicates per-key rate limiting, and 401 indicates authentication trouble

Demon Route currently describes a conditional complimentary lane of up to 50 requests/day after a confirmed deposit, and claims failed requests are refunded. These are provider policies, not features implemented or verified by OmniRoute. Check the official pricing before enabling automatic fallback: OmniRoute does not guarantee refunds or zero cost. A 429 can be transient; a 402 needs balance rather than repeated immediate retries. Its own documentation advertises a 30-day notice and 90-day successor policy for retired IDs, but old IDs are not guaranteed by the local seed list. Privacy statements about third-party upstream hosts are also governed by Demon Route and its hosts.

If model import fails, first check the key and its per-key model entitlement; do not interpret an offline seed fallback as a successful live sync. If chat fails with 404 on an old model, re-import the catalog and choose the current upstream channel name. If streaming stalls, inspect the client timeout and account limits without disabling OmniRoute's normal outbound URL, proxy, or credential safeguards.

Acceptance and troubleshooting checklist

• Before testing: confirm the connection is enabled, its Bearer key belongs to the intended Demon Route account, and the selected model appears in that account's live catalog. Static seeds are not proof of access.
• Discovery: use Import from /models, then check the imported identifiers without changing their case or the vendor/model slash. If the live request fails, distinguish the fallback seed from a successful sync.
• Chat: issue a short non-streaming request first, then verify streaming returns valid SSE events and terminates. Client-side timeouts may need to exceed the built-in 120-second upstream timeout.
• Routing: check a direct model call before adding it to a combo. Then verify fallback behavior and that client-visible errors, traces, and logs do not disclose the stored provider key.
• Errors: troubleshoot 401 by rechecking the key, 402 by checking prepaid balance, 429 by respecting the provider's rate limit, and 404 by re-importing the live model list. Avoid immediate unbounded retries.
• Costs: successful live chat tests can be billable. No live account entitlement, upstream refund, quota, or pricing assertions are established by offline tests in this PR.

For broader provider and combo workflows, see the User Guide.