← Back to all sparks
A

Ably

INFRA · APIS
Velocity6.3

Ably graduates LiveObjects to stable across Java and iOS SDKs, completing its real-time state-sync layer for AI agent deployments.

real-time-infraliveobjectsagentic-aipub-subsdkterraform
Current state
Ably has spent the past two months landing its LiveObjects feature—a path-based API for managing fast-changing shared state—across its major client SDKs. Java SDK v1.8.1 marks the last major platform leaving experimental, following the Cocoa/Swift SDK in early September. In parallel, the AI Transport JS SDK has grown Temporal workflow support and Vercel AI SDK v7 compatibility, and the Terraform provider reached v1.1.0 with significant new surface area including moderation rules and before-publish hooks.
Where it's heading
Ably is building toward becoming the real-time infrastructure layer for agentic AI systems: LiveObjects handles shared state across multiple agents; the AI Transport SDK handles the run lifecycle (turn framing, Temporal orchestration); the new E2E encryption for annotation payloads closes a security gap in the model. The direction is away from general-purpose pub/sub toward targeted infrastructure that connects AI agents to each other and to human-facing interfaces.
Prediction
The next likely move is developer experience work around multi-agent state coordination—LiveObjects is now stable but production wiring patterns are probably still rough. Expect documentation, example patterns, and possibly a managed agent-state abstraction built on top of the stable API.

Recent moves

  1. 5d ago

    Java SDK v1.8.1: LiveObjects Stable Release

    ⚡ SPARK

    LiveObjects exits experimental in the Java SDK—the last major platform to make the transition—completing stable coverage across JavaScript, iOS/Swift, and Java. Production-ready shared-state synchronization via path-based access to LiveMap and LiveCounter structures is now available without experimental caveats across the core platform matrix for AI-agent deployments.

    View source ↗
  2. 7d ago

    Ably Terraform Provider v1.1.0

    The Terraform provider gains data sources for apps, keys, namespaces, and queues—filling the gap that forced teams to combine Terraform with out-of-band API calls. Moderation rules and before-publish hooks are now infra-as-code citizens, and a config-generation tool helps teams migrate hand-built accounts into IaC. This matters most to teams managing Ably at scale.

    View source ↗
  3. 12d ago

    Cocoa SDK v1.4.0: LiveObjects stable release

    LiveObjects goes stable on iOS/Swift ahead of the Java SDK, with breaking API changes that cleaned up the primitive value accessor design before committing to stability. The annotation encryption fix in the same release is a security improvement for encrypted-channel users whose prior annotation payloads were not covered.

    View source ↗
  4. 22d ago

    Ably-js v2.28.0: end to end encryption for annotation data payloads

    End-to-end encryption now covers annotation data payloads in ably-js—previously exempt because server-side aggregation required plaintext access. The aggregation model was redesigned to keep aggregatable metadata outside the payload, enabling encryption without removing the feature. A real security improvement for teams using annotations on encrypted channels.

    View source ↗
  5. 23d ago

    AI Transport JS SDK v0.8.0: Temporal plugin and Vercel AI SDK v7 support

    The AI Transport JS SDK gains a Temporal plugin that handles the full agent-turn lifecycle—opening a run, suspending for client input, cleanup on failure—without boilerplate. Vercel AI SDK v7 compatibility is added alongside v6. Both additions extend Ably's reach into teams building durable, orchestrated AI workflows rather than stateless API calls.

    View source ↗
  6. 26d ago

    Capability semantics changes for `xxx:*` wildcard resources

    A breaking change to capability semantics removes the unintuitive behavior where a wildcard like `ns:*` also granted access to the bare channel `ns`. Teams wanting access to both must now declare both explicitly. A clean-up of the auth model's edge cases—but one that could break existing capability definitions and requires review before upgrading.

    View source ↗