openai_responses_store
Persists Responses API responses to the configured response store backend.
Versions marked “overview” do not contain this page. Selecting one opens that version’s documentation overview.
On this page
Persists Responses API responses to the configured response store backend.
Configuration Notes
For a stored response created with stream: true, GET /v1/responses/{id}?stream=true replays its completed SSE event log. starting_after=N skips events through sequence number N. A plain GET returns the stored response as JSON.
Replay becomes available only after the original response and event log are persisted. A GET before the response is stored returns 404; a stored response without a complete replay log returns 400 for stream=true. This endpoint does not follow generation in progress. If the original foreground connection drops, generation and replay are not guaranteed to complete.
Configuration
| Field | Type | Required | Description |
|---|---|---|---|
backend | sqlite | postgres | yes | Storage backend to use. |
database_url | string (secret) | yes | Database connection URL. Wrapped in [SecretString] to prevent accidental logging of credentials. |
responses_table | string | yes | Table name for response records. |
conversations_table | string | yes | Table name for conversation message records. |
ssl_mode | SslMode | no | TLS mode for PostgreSQL connections. Only valid when backend is postgres. Overrides any sslmode parameter in the connection URL. |
ssl_root_cert | string (secret) | no | Path to a PEM-encoded root CA certificate for PostgreSQL TLS verification. Only valid when backend is postgres and the effective SSL mode is verify-ca or verify-full. |
ssl_client_cert | string (secret) | no | Path to a PEM-encoded client certificate for mutual TLS with PostgreSQL. Only valid when backend is postgres and the effective SSL mode is verify-ca or verify-full. Must be configured together with ssl_client_key. Enables certificate authentication so the server does not challenge for a password. |
ssl_client_key | string (secret) | no | Path to the PEM-encoded private key for ssl_client_cert. Only valid when backend is postgres. Must be an unencrypted PKCS#8 key (mode 0600) and configured together with ssl_client_cert. The native-tls backend (OpenSSL on Linux, Security.framework on macOS) accepts PKCS#8 only; convert a SEC1/PKCS#1 key with openssl pkcs8 -topk8 -nocrypt. |
require_certificate_authentication | bool | no | Enforce the certificate-authentication compliance profile for PostgreSQL. When enabled, the filter fails to start unless ssl_mode is verify-full, both ssl_client_cert and ssl_client_key are set, and no password reaches the connection (rejecting a password in database_url, TLS parameters in database_url, and the PGPASSWORD environment variable). It also rejects non-addressing connection parameters in database_url (application_name, options/options[...], statement-cache-capacity), which the certificate-authentication rebuild would silently drop; set such defaults on the database role instead (ALTER ROLE ... SET ...). The ssl_client_key file must also be owner-only (mode 0600, enforced on Unix). This keeps application-side password cryptography off the connection path. The PostgreSQL server must independently use a cert rule in pg_hba.conf; the proxy cannot enforce that server-side requirement. |
allow_private_database_url | bool | no | Allow PostgreSQL URLs that target local-sensitive addresses. By default, DNS names, localhost, loopback, private, link-local, cloud metadata, unspecified, and Unix socket targets are rejected. This opt-in is intended for local development and tests; cloud metadata, unspecified, and multicast addresses remain blocked when it is enabled. |
pool | PoolConfig | no | Connection pool tuning options. When omitted, sqlx defaults apply (max_connections = 10, idle_timeout = 600s, acquire_timeout = 30s). |
compression | StoreCompressionConfig | no | Optional payload compression for stored JSON columns. When omitted, payloads are stored uncompressed. Reads auto-detect the format, so enabling compression keeps existing uncompressed records readable. |
max_event_count | NonZeroU32 | no | Maximum number of SSE events retained in a streamed response’s replay log (GET /v1/responses/{id}?stream=true). A stream that exceeds this stops event capture for that response, so its terminal event is never recorded and the response becomes non-replayable. The live client stream and the plain JSON record are unaffected. |
max_event_bytes | NonZeroU64 | no | Maximum total payload bytes retained in a streamed response’s replay log. Same over-budget behavior as max_event_count. |
Example
filter: openai_responses_store
backend: postgres
database_url: postgres://praxis:password@db.example.com/praxis
responses_table: openai_responses
conversations_table: openai_conversation_messages
allow_private_database_url: true