Adapter capabilities

This page reconciles the per-adapter compatibility references into one matrix. It is a snapshot of what each protocol front door translates today, not a roadmap. Where a per-adapter page states a boundary, that page is the source of truth; the cells here summarize and link to it.

Each capability cell carries one status:

• Supported — implemented and exercised on that adapter.
• Supported with caveats — works within a bounded surface; the owning reference page states the boundary.
• Not translated — the adapter exposes no surface for the capability; requests for it are rejected or have no equivalent.

Rows one through six carry a status. Rows seven through ten describe the mechanism rather than grade it.

Convex and Native are data-and-compute surfaces. Cloud Functions is a compute-and-trigger surface: its data-capability cells read against the narrow firebase-admin slice callable from inside a function. Firestore, MongoDB, and DynamoDB are data surfaces.

Capability matrix

Capability	Convex	Firestore	Cloud Functions	MongoDB	DynamoDB	Native
Point CRUD	Supported	Supported	Supported with caveats	Supported	Supported	Supported
Explicit queries + pagination	Supported	Supported with caveats	Not translated	Supported with caveats	Supported	Supported
Secondary / composite indexes	Supported	Not translated	Not translated	Supported	Supported	Supported
Live subscriptions / change events	Supported	Supported	Supported	Not translated	Supported	Supported
Multi-document transactions / atomic batch	Supported with caveats	Supported	Not translated	Supported with caveats	Supported	Supported with caveats
Schema validation	Supported	Not translated	Not translated	Not translated	Not translated	Supported
Auth mechanism	OIDC / custom JWT	Bearer-token app auth	Service principal; callable auth	SCRAM-SHA-256	SigV4 signed requests	Local admin token
Tenant binding model	/convex/{tenantId} path	(default) database; auth-scoped	firebase.json app root	Database name = tenant	Access key → one tenant	/api/tenants/{id} path
Default listener / port	Main listener (:8080)	Main listener (:8080)	Main listener (:8080)	127.0.0.1:27017	127.0.0.1:8000	Main listener (:8080) + /ws
Notable hard limits	No file storage, search, or crons	(default) database only; no offline persistence	Firestore document triggers only	Six filter operators; 16 MiB document; loopback only	No provisioned capacity; single stream shard	Loopback default; admin token required

Row notes

• Point CRUD. Convex omits db.replace. Cloud Functions CRUD is the covered firebase-admin DocumentReference get / set / update / delete slice.
• Explicit queries + pagination. Firestore covers the implemented query subset with cursors; nested field-path helpers stay narrow. MongoDB find accepts only eq, ne, gt, gte, lt, and lte and paginates through getMore. Native exposes query and cursor-paginated query endpoints. Cloud Functions exposes no query builder in its covered admin slice.
• Secondary / composite indexes. Convex declares indexes in the table schema. MongoDB exposes createIndexes, dropIndexes, and listIndexes. DynamoDB supports local and global secondary indexes. Native maintains single-field and composite indexes atomically with writes. Firestore and Cloud Functions document no index-declaration surface.
• Live subscriptions / change events. Convex and Native serve reactive query subscriptions over WebSocket. Firestore onSnapshot runs over the WebSocket Listen channel. Cloud Functions delivers Firestore document triggers. DynamoDB Streams expose change records. The MongoDB endpoint rejects $changeStream with CommandNotSupported.
• Multi-document transactions / atomic batch. Firestore covers writeBatch and runTransaction. MongoDB runs multi-statement transactions scoped to one database, failing a conflicting commit with WriteConflict. DynamoDB TransactWriteItems commits all items or none. Convex applies each mutation’s writes as one atomic set with no interactive transaction API. The Native HTTP surface applies each mutation atomically with its index effects but exposes no multi-document batch endpoint.
• Schema validation. Convex and Native offer optional per-table schema validation; a table without a schema accepts any document. Firestore, MongoDB, and DynamoDB document no document-validation surface (DynamoDB defines key schema only).

Additional wire surfaces

Three further surfaces ship today and are documented separately. They are early, bounded surfaces; the notes below describe only what exists now.

• S3. An S3-compatible HTTP listener served by default on 127.0.0.1:9000 (--no-s3 disables it), authenticated with per-tenant signed access keys. See S3 compatibility.
• Cloudflare. Cloudflare-compatible routes and a binding registry mounted on the main listener by default (--no-cloudflare disables them), including a KV binding surface. See Cloudflare compatibility.
• RESP KV. A Redis-serialization-protocol key-value listener started with nimbus kv, bound to loopback 127.0.0.1:6380, with RESP AUTH and a single bound tenant. See KV compatibility.