Roadmap ​

What's planned for TalaDB, roughly in order of impact. Shipped work is recorded in the changelog; this page tracks only what's still open.

Have an idea, or want to help prioritise? Open a GitHub Discussion or a feature request issue.

Developer experience ​

  • Change webhook delivery guarantees — coalescing so a bulk import sends one request instead of hundreds, an opt-in durable outbox for at-least-once delivery, and acknowledged multi-tab write forwarding. See the webhook API.
  • Schema migrations on React Native — the version accessors are wired through the native stack and await on-device verification. Read-time migrations already ship on browser and Node — see Schema Validation.
  • Compound index coverage — use an index when only the leading fields are constrained or the last is a range, per-field descending order, and an array shorthand for createIndex.
  • taladb generate — emit TypeScript interfaces for each collection, inferred from the documents already stored.
  • Svelte and Vue adapters — @taladb/svelte stores and @taladb/vue composables, over the same event model as the React hooks.
  • VS Code extension — filter-expression highlighting, inline document previews, and a collection browser.

The goal is to keep TalaDB among the fastest embedded databases on every JavaScript runtime.

  • Faster vector index builds — graph construction is the current limit on large indexes, and the main thing standing between approximate search and mobile devices.
  • Better approximate-search recall at scale — recall falls off faster as a collection grows than it should.
  • Lower graph traversal cost — smarter cache eviction and leaner cached nodes, so larger graphs stay resident in memory.
  • Wider native SIMD — a runtime-detected AVX2/NEON kernel on top of the portable vectorisation already in place.
  • Faster filtered vector search — stop materialising whole documents just to collect the ids a filter matched.
  • Index tuning guidance by device class — recommended parameters from low-memory phones through desktops.
  • Adaptive cache sizing — size the decoded-vector cache from the device's memory budget instead of one fixed default.
  • Continuous benchmarks — run the suites in CI each release and publish the trend, so regressions are caught before they ship.

Storage ​

  • Pluggable serialisation — swap the internal encoding for MessagePack or CBOR, to interoperate with formats you already use.
  • Document TTL — set an expiry when you write a document and have it swept automatically.

Platform ​

  • Swift and Kotlin packages — first-party wrappers over the C FFI for native iOS and Android apps, without React Native.
  • WASI target — run the same engine inside Wasmtime, WasmEdge and Fastly Compute.