Smart-model routing

Where each smart model is routed right now, live.

Discovered servers

Drag the โ ฟ handle or use โ–ฒโ–ผ to set priority (1 = preferred for proxy routing & failover).
+ Add a server manually
Running in Coolify? To reach another Coolify service by name (e.g. http://ollama), the dashboard container must share its Docker network โ€” or use host.docker.internal / the host's LAN IP if the server's port is published on the host. See the README "Reaching Ollama on Coolify".
0
Proxied requests
0
Succeeded
0
Failed

Top servers (by requests)

Top models (by requests)

Request log

Every proxied request and the server it was routed to. Rows with ๐Ÿ” (last 100) keep the prompt & response โ€” click to view.
TimeClientModelRouted toFormatStreamStatusLatencyTry

Clients

Every client that has reached the proxy, by originating IP + app. Set an alias โ€” it appears as a tag in the Proxy & Logs Client column.
AliasIPAppX-TitleUser agentRequestsAvg req sizeAvg responseLast seen

Models

Every model across your servers, with the latest benchmark per server. Add everywhere pulls on all Ollama servers; Remove everywhere deletes it. (LM Studio / generic OpenAI servers have no download API โ€” those are reported as unsupported.)
ModelServersBestPer-server scores (alias)Actions

Benchmark results

Run benchmarks from the Servers tab. Each runs 3ร— and keeps the best (the first run loads the model into memory); tokens/sec is measured over generation time, excluding time-to-first-token.
TimeServerModelTokens/secTTFTTotalTokensResult

Connect a client

Point any Ollama- or OpenAI-compatible app at this box.

Settings

Account

Signed in as โ€”. This admin account was created on first run.

Proxy API keys

Require a key on the proxy (/api/*, /v1/*). Each key becomes the client name in Clients/Logs. General keeps the client's own X-Title / User-Agent.