What's new for you
Bundled Codex provider with plugin-owned app-server harness for `codex/gpt-*` models; optional Active Memory plugin (dedicated memory sub-agent before replies); experimental MLX speech for Talk Mode on macOS; Seedance 2.0 in fal video tooling; Teams message actions; `openclaw exec-policy` CLI; gateway `commands.list` RPC; Matrix MSC4357 draft streaming markers; broad browser/sandbox/exec security hardening; plus fixes for WhatsApp media, Teams downloads, gateway startup, and many channel routing edge cases.
- Codex — bundled provider + harness so `codex/gpt-*` uses Codex auth, threads, and compaction; `openai/gpt-*` stays on OpenAI path
- Active Memory — optional plugin: sub-agent pulls preferences and context before the main reply (`/verbose`, transcript modes)
- macOS Talk — experimental local MLX speech with interruption and system-voice fallback
- CLI — `openclaw exec-policy` syncs `tools.exec.*` with local exec approvals file
- Gateway — `commands.list` RPC for remote clients to discover runtime commands
Release highlights
| Feature | Details |
|---|---|
| Bundled Codex provider | Adds a bundled Codex provider plus plugin-owned app-server harness so codex/gpt-* models use Codex-managed auth, native threads, discovery, and compaction. Standard openai/gpt-* routes stay on the normal OpenAI provider path. |
| Active Memory plugin | Optional plugin runs a dedicated memory sub-agent before the main model reply so ongoing chats can surface preferences and past context without manual “remember this” prompts. Supports modes, /verbose inspection, and optional transcript persistence for debugging. |
| macOS MLX Talk Mode | Experimental local MLX speech provider for Talk Mode: explicit provider selection, local playback, interruption handling, and system-voice fallback when MLX is unavailable. |
| openclaw exec-policy | New openclaw exec-policy CLI with show, preset, and set keeps tools.exec.* aligned with the local exec approvals file, with conflict detection and safer rollbacks after edits. |
| Gateway commands.list RPC | Remote gateway clients can call commands.list to discover runtime-native, text, skill, and plugin commands with surface-aware naming and serialized argument metadata for custom UIs. |
| Matrix MSC4357 draft streaming | Draft preview sends and edits emit MSC4357 live markers so Matrix clients that support them can show a typewriter-style animation until the final edit lands. |
| Microsoft Teams message actions | Teams gains first-class message actions: pin, unpin, mark read, react, and list reactions so agents match native Teams semantics. |
Feature deep dives
Learn more about the changes that matter most. Bug fixes and minor improvements are listed on GitHub.
Bundled Codex provider & harness
Codex/gpt-* routes use the bundled Codex provider with native threads, discovery, and compaction.
Read more →Active Memory plugin
Dedicated memory sub-agent before the main reply; modes, overrides, and optional transcript debugging.
Read more →macOS MLX speech (Talk Mode)
Experimental local MLX TTS with interruption and fallback to system voices.
Read more →openclaw exec-policy
Align tools.exec.* config with the local exec approvals file via show, preset, and set.
Read more →commands.list RPC
List gateway commands with surfaces and serialized argument metadata for remote UIs.
Read more →Matrix draft streaming (MSC4357)
Live markers on draft sends and edits for animated previews until the final edit lands.
Read more →Teams message actions
Pin, unpin, read state, reactions, and reaction listing for Teams messages.
Read more →Full release notes
The complete changelog with all changes, fixes, and technical details is on the official GitHub release page.
Open v2026.4.10 on GitHub →