Fictura
Error tracking

Logs

Send your backend's log lines to Fictura and read them joined to the trace and the error they belong to.

Search what your backend logged, see volume by level, and open any line to reach its trace. A line logged during a request carries that request's trace, route and user, so an issue and a trace can each show their own lines.

Turn logs on

From version 0.4 of the Python SDK, fictura.init sends them by default. Your existing logging calls at INFO and above arrive, without changing a call.

main.py
import logging
import fictura

fictura.init(
    service_name="my-api",
    logs_level=logging.WARNING,   # optional: only warnings and up
    # enable_logs=False,          # optional: send no logs at all
)

Using the older Fictura(...) constructor instead of init? Pass enable_logs=True.

Logs use their own queue, so a flood of them can't push out an error report, and the SDK never logs its own sending.

Write your own lines

fictura.logger has trace, debug, info, warning, error and fatal. Put {placeholders} in the message: keyword arguments fill them in and become attributes you can read on the line. These lines always go to the Logs page, whatever logs_level says, and never become issues.

Python
fictura.logger.info("Order {order_id} paid", order_id=order.id, amount=1200)
fictura.logger.error("Refund failed", attributes={"order_id": order.id, "reason": reason})

To edit or drop a line before it leaves, use before_send_log:

Python
def before_send_log(line):
    if line.get("logger") == "uvicorn.access":
        return None   # drop it
    return line

fictura.init(before_send_log=before_send_log)

Read them

The Logs page, under Reliability:

  • Search the message text, and filter by level, logger, route, release and environment.
  • Choose a time range, up to the 14 days lines are kept.
  • See volume by level over that range.
  • Group by logger, level, route or release to see what is noisiest. Click a row to filter to it.
  • Open a line to read the whole message and its attributes, open its trace, or see other lines from the same logger, route, release or user.
  • Export what matches as CSV or JSONL, up to 10,000 lines.
  • Auto-refresh reads again every 10 seconds while the tab is in view.

An issue's page shows Logs from this trace for its latest event, and a trace's page shows them beside the waterfall. Your coding agent can read them with the search_logs MCP tool.

Limits

Kept for14 days
One messageCut at 8 KB
One line's attributesCut at 4 KB
One request500 lines and a 5 MB body
Rate30,000 lines a minute per app. Over that the API answers 429 with Retry-After.
A line outside the kept windowRejected, not stored
SDK queue5,000 lines, dropping the oldest. Sent in batches of up to 500.

Searching the text is a substring match on the message, always inside the chosen time range. A very wide search can time out: narrow the range or add a filter.

From another language

Send lines to POST /api/v1/logs with your SDK app key in x-api-key:

Request body
{
  "logs": [
    {
      "ts": "2026-11-05T12:00:00Z",
      "level": "info",
      "message": "Order 42 paid",
      "logger": "app.billing",
      "template": "Order {order_id} paid",
      "trace_id": "4bf92f3577b34da6a3ce929d0e0e4736",
      "span_id": "00f067aa0ba902b7",
      "route": "/api/v1/orders",
      "method": "POST",
      "user_id": "user_123",
      "release": "2026.11.1",
      "environment": "production",
      "service": "my-api",
      "server_name": "web-1",
      "attributes": { "order_id": 42 }
    }
  ]
}

Only message is needed. level is one of trace, debug, info, warning, error or fatal, and info when left out. A line with no ts gets the time it arrived. The answer is { "ok": true, "accepted": 1, "rejected": 0 }: one bad line never costs the rest. Secrets in the message and route are masked when they arrive.

Log lines can hold personal data. See Privacy.

On this page