Data retention #

Hypersnap stores data in three distinct tiers, each with its own retention characteristics:

1. Live message state #

Casts, reactions, follows, verifications, user data — everything that has a Farcaster CRDT behind it — is subject to the per-FID storage limits defined by the network. See /v2/farcaster/user/storage-allocations and /v2/farcaster/user/storage-usage for the live state of a given FID.

When an FID's storage expires or is exceeded, older messages are pruned according to the store-specific CRDT rules. Hypersnap doesn't keep a local archive beyond what the protocol preserves.

2. Webhook retry queue #

Failed deliveries are held on a durable RocksDB-backed retry queue for the duration of the retry schedule:

• Default retry_max_attempts = 5
• Default retry_initial_backoff_ms = 500 (doubles each attempt)
• Total retry window ≈ a few minutes for a 5-retry sequence

If all retries fail, the delivery is dropped. It is not held indefinitely.

3. Mini-app notification tokens #

Notification tokens are stored per (fid, app_id) indefinitely until:

• The Farcaster client sends a notifications_disabled or miniapp_removed JFS event (token is deleted / disabled).
• The send endpoint receives an invalidTokens response from the client for that token (Hypersnap deletes it).
• The mini app itself is deleted.

There is no background TTL sweeping the token store.

4. Nonce LRU #

EIP-712 nonces are held in-memory for 2 × signed_at_window_secs (default 10 minutes). This is long enough to prevent replay inside the signed-at window and short enough to bound memory. Replay attempts outside the window are rejected by clock skew regardless.

What's not stored #

• No session state, API keys, or user-facing credentials.
• No audit log of read traffic.
• No mirror of the operator's reverse-proxy logs.

Hypersnap is a Farcaster node first and a developer API second. Long-horizon retention beyond what the Farcaster protocol preserves is your responsibility if you need it — wire up an agent to a webhook, mirror the events you care about, and own that data in your own system.