Database and product milestones, linked to their behavior and evidence. These are development updates; they do not imply a stable release or completed production acceptance.
LET and nested FOR blocks bind typed local values into the shared query planner. SELECT sources keep native projection and aggregate values without a JSON round trip. Shared work limits, rollback, stale-source validation and original receipts pass functional checks. The earlier 8141f00 comparison now retains all 80 FZ trials; loop performance, fluent APIs and correlated indexes remain open.
Read the scoped program contract →QQL programs return ordered statement results through the shared transaction engine. Bulk INSERT accepts bound document arrays; partial object updates preserve nested siblings, with explicit replacement and array edit functions. Equivalent secondary indexes reject before another build. Functional recovery and client checks pass; the FZ regression comparison and general FOR/fluent language work remain open.
Use programs and partial updates →The standalone WebSocket API adds bounded concurrent requests through the same authentication and transaction path as HTTP. HTTP keep-alive is verified. The shared TS/JS client and Studio support a separate transport choice for each instance, with explicit resolution of interrupted writes. FZ transport component measurements are retained; full database, TLS and load-balancer performance acceptance remains open.
Use the persistent protocol →New databases default to authenticated plaintext storage, with explicit encryption opt-in for native or external custody. WAL, pages, checkpoints, objects and backups share the selected mode. Studio shows the mode for each connected instance. Verified mode conversion remains in development.
Read the protection model →Unordered aggregates share field resolution and retain accumulators instead of input rows. Bounded sorts resolve keys once and preserve stable ties while selecting the requested window. Correctness checks and FZ comparisons retain both gains and regressions; full workload performance acceptance remains open.
Follow the execution path →The public website and Studio now use distinct development addresses and build identities. The server rejects website assets in its Studio slot. Studio adds snapshot inventory, capacity and worker settings, direct collection index access, atomic index replacement and offline maintenance guidance.
Read the management API →The management console moves to studio/ with its existing history. The public app/ introduces documentation, architecture, benchmark tables and generated HTTP reference. Client downloads use the same first-party schema and validators.
Explore the documentation →Private index construction now applies bounded batches through the persistent-tree mutation primitive. The change retains 28 index publication fault cases. Earlier laptop timing claims have been withdrawn; the benchmark page now uses checked FZ server results.
Read the measurements →Index construction and expanded predicates use selected-field views to avoid decoding unrelated document payloads. This builds on the common execution context, indexed candidate paths and exact residual checks.
Follow query execution →Stopped-instance rewrite and offline backup/restore share the engine generation lifecycle and configured native or external custody. Retained fault evidence covers logical and physical stores, receipts and exact recovery.
Understand file maintenance →The first retained comparative baseline checks embedded Qirava/SQLite and Qirava/ArangoDB HTTP in separate lanes. It records working point reads and significant durable-write and complex-query gaps, creating an explicit baseline for the next changes.
Explore the baseline →
DATABASE