Archive logs in your own bucket
Observer writes the previous UTC day, per project, to S3 or Google Cloud Storage. Events are archived as stored, after redaction. Credentials stay encrypted in the e-volv vault and are never returned to the browser.
Objects and completion
Each part is gzip NDJSON with at most 100,000 events. The object path is prefix/project/YYYY/MM/DD/part-NNNN.ndjson.gz. After every part is written, Observer writes _SUCCESS with the part and event counts. An existing success marker skips the day. Without one, the next run starts at part 0000 and overwrites partial files. Three failed days disable the target and notify workspace owners.
Test and read
Use Test connection in Observer Usage to write and delete one small object. The settings page shows recent runs and sanitized errors. DuckDB can query the compressed newline-delimited JSON directly:
SELECT ts, project_id, severity, message FROM read_ndjson_objects( 's3://my-bucket/prod/log_project:api/2026/11/18/*.ndjson.gz', compression = 'gzip' ) WHERE severity >= 17 ORDER BY ts DESC LIMIT 100;
Restore is not included. Keep your normal bucket lifecycle and access policies in place; archive export does not change Observer retention.