OpenClaw 2026.4.26 went live on April 28 with a sprawling changelog. The headline items: a generic browser realtime transport contract for Talk (initially wired up for Google Live browser sessions with constrained ephemeral tokens and a Gateway relay for backend-only voice plugins), an openclaw matrix encryption setup CLI flow, Cerebras as a bundled provider, model-specific Ollama embedding prefixes, and a new openclaw migrate command with bundled Claude and Hermes importers. This is a straight read of the release notes — no operator-thread hearsay — with editorial commentary on what matters.
The most-discussed bullet is the new browser realtime transport contract under Control UI/Talk. Per the release notes, this is a generic contract — not a specific config switch — and the first concrete consumer is Google Live browser Talk sessions, which run with constrained ephemeral tokens. Backend-only realtime voice plugins reach the browser through a Gateway relay rather than punching out directly.
A related fix in the same release tells you how cautious the team is being: Google Live browser sessions are kept on the WebSocket transport instead of falling back to WebRTC, browser Google Live WebSocket endpoints are validated, and Gateway relay sessions are capped per browser connection. Stale browser-native voice buttons that bypassed the configured Talk/TTS provider were also removed.
For OpenClaw operators: this is foundation work, not a finished operator-facing feature. Expect future Talk plugins to land on this contract first.
openclaw matrix encryption setup is new. The release notes describe it as a single setup flow that enables Matrix encryption, bootstraps recovery, and prints verification status. If you run any agents through Matrix as the message bus, this consolidates what used to be multiple manual steps. Credit to @gumadeiras in the release notes.
Cerebras lands as a bundled plugin with onboarding, a static model catalog, docs, and manifest-owned endpoint metadata. This is part of a broader pattern in 2026.4.26: pre-runtime model-id normalization, provider endpoint host metadata, and OpenAI-compatible request-family hints have all been moved into plugin manifests, so core no longer carries bundled-provider routing tables. Adding a new provider is now a manifest exercise, not a core patch.
Memory-search queries against Ollama embedding models now apply model-specific retrieval query prefixes for nomic-embed-text, qwen3-embedding, and mxbai-embed-large. Document batches are left unchanged. If you have memory recall issues against any of those three models post-upgrade, the prefix change is the first place to look.
There is also a sibling change for OpenAI-compatible memory endpoints: optional memorySearch.inputType, queryInputType, and documentInputType config for asymmetric embedding endpoints, including direct query embeddings and provider batch indexing.
Two CLI/migration bullets are worth flagging together:
openclaw migrate itself, with plan, dry-run, JSON output, pre-migration backup, onboarding detection, archive-only report copies, and a bundled Hermes importer for configuration, memory/plugin hints, model providers, MCP servers, skills, and supported credentials.If you are coming to OpenClaw from Claude Code, Claude Desktop, or Hermes, the migration path is now an in-tree command with a real dry-run mode rather than a doc page of manual steps.
A non-exhaustive sample of the rest:
agents.defaults.compaction.maxActiveTranscriptBytes preflight that triggers normal local compaction when the active JSONL grows too large, then rotates the transcript so future turns land on a smaller successor file.OPENCLAW_PLUGIN_STAGE_DIR can now contain layered runtime-dependency roots, resolving read-only preinstalled deps before installing missing deps into the writable root.openclaw doctor --fix can quarantine the offending plugin instead of letting every channel crash-loop.none/default or off/on instead of OpenAI-only low/medium.openclaw matrix encryption setup collapses Matrix encryption, recovery bootstrap, and verification status into one CLI flow.nomic-embed-text, qwen3-embedding, and mxbai-embed-large; document batches are unchanged.openclaw migrate adds a real dry-run-capable migration command with bundled Claude Code, Claude Desktop, and Hermes importers — the supported way to move existing setups onto OpenClaw.Discover more content: