Request log
Query and export the EUrouter request log: who made each request, which model and country processed it, the decision, its reasons and the cost.
EUrouter writes one log row for every inference request, also for a request it blocked. A row says who made the request, which model and provider served it, in which countries it was processed, what it cost, and the decision EUrouter took with its reasons. It never holds a prompt, a completion or any other content.
Use the log to answer an auditor, to feed a SIEM, or to find out why one request was refused.
Who sees what
- Owners and admins see every request of the organization.
- Members see the requests made with their own API keys or while signed in.
Call the log API with an organization-wide management API key or signed in. An API key for model requests and a management key bound to one workspace get a 403.
In the dashboard
The Logs page in the dashboard shows the log of the workspace you have open, newest first, 50 rows a page. Each row has the time in UTC, the API key, the model with its provider and country, the status and the reason. Click a row for its details: attempts and fallback, timings, the decision with every reason, personal data counts, prompt-injection findings, routing, tokens, cost and a request id to copy.
Filter by period (up to 31 days in the list), status, reason and model. Owners and admins can also filter by member and switch to every workspace of the organization. A request's details offer Only this key to filter by its key. The filters live in the page address, so you can bookmark or share a view.
Export CSV and Export JSON download the same filters for the period you picked, with no 31-day limit.
Look up one request
Every inference response carries an x-request-id header, also when the request was blocked. Pass it to
GET /logs/{request_id}:
curl "https://api.eurouter.ai/api/v1/logs/5f0c2a4e-8f1b-4c7e-9a53-2f6f0d1c9b7a" \
-H "Authorization: Bearer $EUROUTER_MANAGEMENT_KEY"{
"data": {
"id": "5f0c2a4e-8f1b-4c7e-9a53-2f6f0d1c9b7a",
"created_at": "2026-11-03T09:15:00.120000Z",
"key_name": "Support bot",
"member_email": "dana@example.com",
"requested_model": "openai/gpt-4o",
"model": null,
"provider": null,
"processing_countries": [],
"decision": "blocked",
"reason_codes": ["policy.eu_owned_only"],
"reasons": [{ "code": "policy.eu_owned_only", "sentence": "Your rule: EU-owned providers only" }],
"pii_counts": {},
"prompt_injection": null,
"http_status": 403,
"cost_eur": null,
"addon_charges": [{ "addon": "prompt_injection_ai_check", "units": 120, "cost_eur": "0.0000240000", "free": false }]
}
}cost_eur is what the model cost. Paid add-ons that ran on the request come on top, each as its own line in
addon_charges: the add-on (prompt_injection_ai_check for the AI check by Mistral),
what it measured in units (for the AI check, the tokens checked), its cost_eur and whether the monthly free allowance
of your plan covered it (free). A request the AI check blocked has no model cost but can have an AI check line. The
list is empty when no add-on ran.
The row is written right after the response ends, so a lookup in the same second can return a 404. Try again a moment later. A request you may not see also returns a 404.
List requests
GET /logs returns the requests of a period, newest first. from and to are required and a call covers at most 31
days.
curl "https://api.eurouter.ai/api/v1/logs?from=2026-11-01&to=2026-12-01&decision=blocked" \
-H "Authorization: Bearer $EUROUTER_MANAGEMENT_KEY"| Parameter | What it does |
|---|---|
from | Start of the period, inclusive. An ISO 8601 date or date-time; without a time zone it is UTC |
to | End of the period, exclusive. to=2026-12-01 includes all of November |
key_id | Only requests made with this API key |
workspace_id | Only requests made in this workspace |
member_id | Only requests made by this member |
model | Only requests served by this model, or that asked for it when no model served them |
provider | Only requests served by this provider |
country | Only requests processed in this country, such as FR |
decision | allowed, redacted, blocked or error |
reason_code | Only requests that carry this reason code |
limit | Rows per page, 1 to 100. The default is 50 |
cursor | The next_cursor of the previous page |
The response has the rows under data and a next_cursor. Pass it as cursor, with the same other parameters, to get
the next page. On the last page next_cursor is null.
Decisions
| Decision | What happened |
|---|---|
allowed | Sent to a provider unchanged |
redacted | Sent to a provider after personal data was redacted |
blocked | Refused before it reached a provider |
error | Failed with an error |
Reason codes
A row carries a reason code for every rule that blocked, redacted or flagged the request. A request that passed every
rule has none. The API returns each code with its sentence under reasons.
| Reason code | Meaning |
|---|---|
residency.countries_unknown | EUrouter rule: verified processing countries only |
residency.country_not_allowed | EUrouter rule: processing inside the allowed countries only |
policy.country_not_allowed | Your rule: allowed processing countries |
policy.eu_owned_only | Your rule: EU-owned providers only |
policy.certification_missing | Your rule: required certifications |
policy.keeps_data | Your rule: no providers that keep data |
policy.trains_on_data | Your rule: no providers that may train on your data |
policy.provider_not_allowed | Your rule: allowed providers |
policy.model_not_allowed | Your rule: allowed models |
pii.blocked | Your rule: block requests that contain personal data |
pii.scan_limit | The request was too large to scan for personal data |
pii.redacted | Personal data was redacted before the request was sent |
pii.flagged | Personal data was found and flagged |
prompt_injection.blocked | Your rule: block requests that contain a known prompt-injection phrasing |
prompt_injection.flagged | A known prompt-injection phrasing was found and flagged |
prompt_injection.scan_limit | The request was too large to check completely for prompt injection |
budget.key_cap_reached | The API key reached its spending limit |
budget.org_cap_reached | The organization reached its monthly budget |
budget.workspace_cap_reached | The workspace reached its monthly budget |
budget.member_cap_reached | The member reached their monthly budget |
balance.insufficient | The account balance was too low |
plan.rate_limited | The rate limit of your plan was reached |
plan.monthly_quota | The request limit of your plan was reached |
model.not_in_eu | The model is not available from an EU provider |
model.not_found | The model does not exist or is retired |
request.invalid | The request was not valid |
Export
GET /logs/export downloads the log of any period as one file, with the same filters as the list. There is no 31-day
limit: the file is streamed, so a full year works too.
curl "https://api.eurouter.ai/api/v1/logs/export?from=2026-01-01&to=2027-01-01&format=csv" \
-H "Authorization: Bearer $EUROUTER_MANAGEMENT_KEY" \
-o request-log-2026.csvformat=csv (the default) opens in a spreadsheet. format=json returns one JSON object per line (NDJSON), each in the
shape of a row from the list.
CSV columns
| Column | Content |
|---|---|
request_id | The x-request-id of the request |
time | When EUrouter received the request, ISO 8601 in UTC |
key_name | Name of the API key at the time of the request |
key_id | Id of the API key. Empty for a request made signed in |
member_email | Email address of the member who owns the key or was signed in |
member_id | User id of that member |
endpoint | API the request called, such as chat.completions |
requested_model | Model the request asked for |
model | Model that served the request. Empty when no model served it |
provider | Provider that served the request. Empty when it reached no provider |
processing_countries | Countries where the provider processes the model, such as FR; NL |
decision | allowed, redacted, blocked or error |
reason_codes | Reason codes, such as policy.eu_owned_only; pii.flagged |
reasons | The same reasons as sentences |
pii_counts | Personal data items found per type, such as EMAIL:2; PHONE:1 |
prompt_injection_action | The prompt-injection rule that applied, flag or block. Empty when nothing was found and every input was checked |
prompt_injection_count | Number of known prompt-injection phrasings found |
prompt_injection_techniques | Techniques found, such as instruction_override; role_switch, or ai_jailbreak for a finding of the AI check |
prompt_injection_not_scanned | Inputs that could not be checked, such as image_not_scanned, or ai_check_unavailable and ai_check_not_allowed |
tokens_in | Prompt tokens billed |
tokens_out | Completion tokens billed |
cost_eur | Amount charged, in EUR |
addon_charges | Paid add-ons, each as add-on, units and cost, or free, such as prompt_injection_ai_check:120:0.0000240000 |
http_status | HTTP status of the response |
latency_ms | Time from request to the end of the response, in milliseconds |
images_in | Input images for a billed image request, when available |
images_out | Completed images returned and billed, when available |
A cell with several values separates them with ; . A text cell that starts with =, +, - or @ gets a leading
apostrophe, so a spreadsheet shows it as text and never runs it as a formula.
See the Logs API reference for every field.