We use docker. You can run it without but it's not recommended.
cp .env.example .env # and fill them in obviously
docker compose up -d --buildCompose stores the SQLite database and local uploads under ./data/, which Docker creates on first start. Back up that directory if you use local storage. To keep runtime data elsewhere, change the two bind-mount paths in docker-compose.yml.
- Go
- SQLite
- Cloudflare R2 (optional)
- Docker
- Anonymous uploads
- No persisted IP logging
- Public uploads default to a 256 MiB maximum and two upload requests per IP per minute (
ANONHOST_MAX_FILE_SIZE,ANONHOST_RATE_LIMIT); direct-to-R2 initialization uses the same limiter, and finalization requires a short-lived server-issued upload intent plus an exact object-size check - Upload processing is capped at three concurrent requests; excess requests receive 503 and can retry
- File deletion via cryptographic tokens
- Optional file expiration
- Direct-to-R2 uploads via presigned URLs (faster uploads through the webui)
- Archive previews with tree navigation
- CLI tool
- ShareX support
- Maintenance mode (
ANONHOST_MAINTENANCE): rejects every upload endpoint with 503 while it is on; already uploaded files stay downloadable and the web UI shows a notice - Automatic retention: files older than 14 days (
ANONHOST_RETENTION_DAYS) are wiped (storage object + DB row) by an hourly background job;expires_atfiles are wiped too - Deduplication: each cleanup ends with a content-hash pass over plaintext files. Byte-identical uploads share one master copy (duplicate objects deleted, duplicate IDs repointed, all URLs/deletion tokens keep working); deletes are refcount-aware so shared bytes survive until the last ID is gone. Hash backfill is budgeted per run (
ANONHOST_DEDUP_MAX_FILES/ANONHOST_DEDUP_MAX_BYTES, smallest first) and runs on a client-disconnect-proof context, converging over successive runs - Orphan prune (manual only,
POST /api/orphanson the admin port,?dry_run=1to preview): drops DB rows whose storage objects are gone, e.g. after manual R2 deletes. Starts a background job (202, pollGET /api/orphans/status); only 404-class misses are pruned, transient errors are skipped, and runs aborting when over half the rows look orphaned - Internal admin portal on a separate bind (
ANONHOST_ADMIN_BIND, default127.0.0.1:1985): stats, search, delete, manual cleanup, archive rescan. Protect withANONHOST_ADMIN_KEYand keep it off the public interface.
For a public instance, put Cloudflare in front of the origin, block direct public access to the origin ports, and enable edge rate limiting or a managed challenge on POST /api/upload and POST /api/upload/direct. Set the edge request body limit to match ANONHOST_MAX_FILE_SIZE plus multipart overhead. App-level per-IP limits are a useful backstop, not a substitute for an edge rule against distributed traffic. Reconcile unregistered R2 objects separately; applying a simple age-based lifecycle to the bucket can delete objects still referenced by newer deduplicated rows.
curl https://anonhost.cc/install | bash
anonhost.sh upload image.png
anonhost.sh upload secret.txt --encrypt # encrypted; key is appended to the URL fragmentSee roadmap for planned improvements.
MIT