OpenClaw v2026.7.2-beta.6: Models and providers, interface notes

v2026.7.2-beta.6 changes a surface that users can see. The release record describes a channel, client, interface, or browser change; this article follows the state behind the screen instead of treating a new button as proof that the workflow is complete.

Because this is a pre-release, use a disposable Gateway first and leave the stable version available for comparison. The release body cannot certify every local combination of provider, channel, platform, and account; use the bounded test below for the path you actually run.

State that a reader can see

State safety and recovery

What changed. protect persisted data with a quarantine store that survives primary-database damage, crash-recoverable SQLite snapshots, crash-durable filesystem publication, schema-upgrade data-loss rejection, and rollback-writer snapshot recovery.

Check it with. For State safety and recovery, inspect state rather than only the new control. Repeat the flow after reload and reconnect, and note the session, pending action, URL, and visible error.

Durable channel delivery

What changed. keep accepted messages recoverable across gateway restarts and local crashes through the shared ingress drain and dead-letter recovery, covering Telegram, Signal, Slack, QQBot, Twitch, Synology Chat, Tlon, IRC, and Zalo User. #108656, #107246, #109911.

Check it with. Use Durable channel delivery as a routing test: capture the destination before the send, interrupt the transport, and verify that retry logic did not duplicate or redirect the response.

Session rewind and branching

What changed. rewind or fork conversations from individual messages, switch transcript branches across web and native apps, fork upstream Codex sessions, preserve branch-safe queued sends, reject stale-pane writes, and restore prompt images after a fork.

Check it with. Use Session rewind and branching as a routing test: capture the destination before the send, interrupt the transport, and verify that retry logic did not duplicate or redirect the response.

Interactive MCP Apps and dashboards

What changed. host ticketed MCP Apps with bound tools, resources, and bounded context updates; open them from channel replies, pin them to durable dashboards, harden their shared sandbox, and let native plugins declare them directly. #109851, #110451, #113218.

Check it with. The useful evidence for Interactive MCP Apps and dashboards is the refusal path as well as the success path. Use synthetic credentials and record the policy that made each decision.

The wiring behind the screen

Questions and approvals everywhere

let agents ask structured questions with option cards across web, channels, macOS, and native apps, while approvals gain push notifications, history, fair queuing, headless resolution, Claude tool-request relay, reviewer detail, and clearer formatted prompts. #85954.

Treat Questions and approvals everywhere as a boundary change. Exercise one permitted request and one refused request with a test identity, then keep both decisions in the log.

Models and providers

add Claude Opus 5 across catalog and runtime, Kimi K3, and GPT Live realtime support with the supported Platform API authentication path. (#113391, #113392, #113633; #113909, #113354.

Read Models and providers as an authorization checkpoint: prove who may act, what the action can reach, and how the rejected attempt is surfaced to the operator.

Local inference and setup

detect local inference providers during onboarding, add in-process llama.cpp GGUF inference and Baseten Model API support, discover models from live provider catalogs, and offer model downloads from web and macOS setup.

The safe reading of Local inference and setup is operational: prove the model can answer the intended task, then force a bounded failure and verify that no side effect is repeated.

A bounded check for this version

1. Start with a disposable checkpoint: record the version, host, provider, and workspace tied to State safety and recovery.
2. Check the release's migration notes and make a restorable backup before changing persistent state.
3. Use a disposable conversation for one send, one reconnect, and one retry; keep the destination identifiers with the result.
4. Leave the channel disabled after the test if its credentials or pairing were created only for this rehearsal.
5. If the result differs from the release body, stop the rollout and attach the log, screenshot, or transcript to the version record.

openclaw --version
openclaw gateway status
openclaw security audit

Close the loop with evidence

A useful maintenance record names the changed behavior, the behavior that stayed stable, the identity that approved the test, and the step that would undo it. Preserve that record beside the release tag.

Edge cases in this tag

Security and authorization. prevent channel allowlists from granting owner access, keep session exports inside the workspace, close a forged-marker/web-search boundary bypass, prevent non-owner ACP session exposure, reject unsafe explicit approval IDs, harden secret redaction and exec/OAuth approvals, validate downloaded install scripts, and prevent insecure secrets-plan writes. Record the failure that this correction addresses before declaring the upgrade complete.

SQLite and data safety. commit session indexes before transcript eviction, preserve state through maintenance races and live-WAL verification, reject invalid backups and schema data loss, make snapshot publication crash-recoverable, retain complete backups after interrupted commits, and evict only the exact corrupted cached database owner so repairs recover without a Gateway restart. This is the edge case to include in a regression check, especially when the release is pre-release.

The authoritative record is the v2026.7.2-beta.6 GitHub release. The release index gives the surrounding sequence, while the channel routing guide supplies the operational background.