Feature · Debug console

An LLM API debug console for models and routes

Before you hand a new model or route to Claude Code or Codex, try it yourself. The Debug console sends real requests through the AutoJev gateway and shows exactly what happened: which model was picked, how long it took, how many tokens it used and the raw response.

Conversation mode and cURL mode
Switch between Chat Completions, Responses and Messages
Up to 4 attachments, 4 MB each
Selected model, timing, tokens, cost and raw JSON

Download AutoJev Star on GitHub

Debug

autojev/quality

Summarize the failing test and propose a fix.

The test fails because the date parser ignores the timezone offset. Normalize to UTC before comparing…

Modelopenrouter/claude-sonnet

First token842 ms

First byte / total3.9 s

Tokens in / out1,204 / 318

Est. cost$0.0084

Test through the same path your agents use

Most API playgrounds talk to one provider directly. That tells you nothing about your route, your protocol conversion or your failover setup. The AutoJev Debug console sends requests to the local gateway, so they go through exactly the same routing, conversion and resilience logic as an agent request — and appear in Logs like any other.

Pick any model or route from a searchable list, choose the API type to send, and start testing. Live tests call your real providers, so they may incur upstream costs.

Two modes

Conversation mode

Chat with a model or route like you would in an app. Multi-turn history is kept until you clear the conversation, which starts a new session. Enter sends, Shift+Enter adds a new line.

cURL mode

Paste or edit a cURL command aimed at the local gateway and execute it. See the HTTP status, response headers and output, with a one-click Format JSON. Stop a streaming request at any time.

A cURL request the console accepts

cURL mode accepts POST requests to /v1/chat/completions, /v1/responses or /v1/messages on the local gateway. The command is parsed, never executed as shell code.

Debug → cURL mode

curl -N -X POST \
  --url 'http://127.0.0.1:9527/v1/responses' \
  -H 'Content-Type: application/json' \
  --data-raw '{
  "model": "autojev/quality",
  "input": "Explain this regex: ^[a-z0-9_-]{1,64}$",
  "stream": true
}'

What every response shows

Detail	Why it helps
Model / route and provider	Confirms which real model a route selected
Routing source and decision reason	Shows whether Jev, local selection or load balancing decided — and why
Duration and first byte	Spot slow providers before an agent does
Input / output and cache tokens	Check that usage is reported correctly
Estimated cost	Based on the prices configured for that model
Upstream attempts	See failover in action, attempt by attempt
Request JSON and raw JSON response	Inspect exactly what was sent and what came back

Built for realistic tests

Switch the API type to test protocol conversion — send Messages to a Chat Completions model, or Responses to a Messages model
Attach up to 4 files under 4 MB each: images, PDFs, or UTF-8 text and code files; paste images straight from the clipboard
Add custom request parameters by name and value
Switch to cURL mode and the current conversation is prefilled as a cURL request you can edit, run or copy

Where the console fits in your workflow

Use it when you add a provider, to confirm a model really answers with tools or images before an agent depends on it. Use it when you change a route, to see which candidate the new rules pick. And use it when an agent misbehaves, to replay a similar request and compare the raw JSON against what the agent expected.

Every debug request is logged like any other, so you can open it later in Logs and compare it with real agent traffic side by side.

Validate a new route in three steps

1
Select the route
Open Debug and search for your route, for example autojev/fast.
2
Send a request your agent would send
Choose the API type your agent uses — Responses for Codex, Messages for Claude Code — and send a prompt, with tools or images if needed.
3
Read the details
Check the selected model, decision reason, timing, tokens and cost. If something looks off, open the raw JSON.

Frequently asked questions

How do I test an LLM API endpoint locally?

Add the provider and model in AutoJev, then open Debug, select the model and send a message. The request goes through the local gateway at http://127.0.0.1:9527 to your provider.

Can I test a route instead of a single model?

Yes. Select any autojev/<route-id> and the console shows which model the route picked and the decision reason.

Do debug requests cost money?

Yes. They are real requests to your providers with your keys, so normal upstream pricing applies.

What can I attach?

Up to four files, each under 4 MB: images (PNG, JPEG, WebP, GIF), PDFs, or UTF-8 text and code files. Whether the model accepts them depends on the upstream model.

Can I run arbitrary cURL commands?

No. cURL mode parses the command and only accepts POST requests to the current local AutoJev gateway on a supported model endpoint. Shell operators and variables are not executed.

Keep exploring

FeaturesUsage & request logsTokens, latency and estimated cost for every request, broken down by agent, model and provider. Metadata only, stored locally.FeaturesProtocol conversionConvert between Chat Completions, Responses and Anthropic Messages in all six directions, with streaming and tool calls.Use casesFor evaluating modelsMeasured latency and tokens/s, a debug console for identical prompts, and per-model stats from real agent traffic.FeaturesModel speed testsMeasure first-content latency and output tokens per second on your own network, on a schedule, and route by the results.FeaturesProvider managementAdd OpenAI, DeepSeek, OpenRouter, Ollama or any compatible endpoint once, or import from CC Switch and Termany.

Updated 2026-10-14

Give every agent the right model.

Install AutoJev, add the providers you already pay for, and connect your agents in one click. Everything stays on your machine.