Engine checks¶
When the provider client starts your engine, it checks what the engine can do and measures how many requests it can serve at once. Flockadoo routes on these results: a request that needs a capability goes only to flocks that passed it.
You can see the results on the flock page:
| Where | What you see |
|---|---|
| Overview tab, Engine checks card | Each capability's result, whether the model is verified, and how many requests the flock takes at once |
| Checks tab | The full record: the model and its build, what each result means for you, and every benchmark level |
The results are what your own client reported. The owner and the workers of a flock can both see them.
Open the Checks tab¶
- Click Flocks in the sidebar.
- Open the flock.
- Open Checks, or click Details on the Engine checks card.
Capabilities¶
| Capability | Requests that need it |
|---|---|
| Tool calling | Requests that send tools |
| Forced tool choice | Requests whose tool_choice names a function or is "required" |
| Reasoning | Every request for a model that thinks (see below) |
| JSON schema | Requests with response_format type json_schema |
| JSON mode | Requests with response_format type json_object |
| Logprobs | Requests that ask for logprobs |
Each capability shows one result:
| Result | Meaning |
|---|---|
| Supported | The engine passed the check. You get requests that need it. |
| Not supported | The engine answered, and it cannot do this. Requests that need it go to other flocks. |
| Not for this model | The model's build does not allow it. For example, a model that does not think has nothing to check for Reasoning. |
| Couldn't check · retrying | The check timed out or the engine errored. The client tries again while the GPU is idle. |
| Checking | The client is checking it now, after the engine or model changed. |
| Not checked yet | The client has not reported this check yet. |
Until a capability is Supported, requests that need it go to other flocks. Ordinary requests keep coming in the meantime.
Warning
Reasoning is stricter than the others. A model that thinks (Qwen3, DeepSeek-R1, gpt-oss, Gemma 4) is served only by flocks whose engine sends the thinking apart from the answer. Until Reasoning is Supported, your flock gets no requests for that model at all.
An embedding model gets no chat checks. The tab says Chat checks do not apply to an embedding model.
Model¶
The Model panel shows the model the flock serves now:
- Verified means the model is an approved build and its weights header matched. Your own price applies only after this. See Model pricing.
- Tool parser and Reasoning parser come from the approved build. A model without a tool parser cannot take tool requests, so Tool calling and Forced tool choice show Not for this model.
Other models your client has registered on the flock are listed below, with whether each is verified.
Concurrency benchmark¶
The client sends 1, 2, 4, 8 and then 16 requests at once. It stops at the first level that fails or gets too slow, and takes the highest level that held up. The flock then takes that many requests at once.
| Result | Meaning |
|---|---|
| Passed | This level held up. |
| In use | The number of requests the flock takes at once. |
| Did not pass | The level where the client stopped. |
For example, flockadoo-flock1 shows 8 requests at once when 8 held up and 16 did not.
If even one request at a time failed, nothing was measured. The card then says Not measured, and the flock takes a starting number until a benchmark completes.