On August 30, 2026, the OpenClaw Foundation released OpenClaw 2.0, the largest update in the project's history. The release bundles over 16,000 pull requests from 933 contributors, including 569 first-time contributors.

It touches every part of the open-source personal AI assistant: installation, messaging, memory, skills, models, automations, browser and native apps, plugins, and security.

The version, tagged v2026.8.1, arrives after nearly seven weeks without a shipment — an unusual pause for a project that had pushed 106 releases in 230 days. Hannes Rudolph wrote on the OpenClaw blog that the team's growth outpaced both the codebase foundation and the shipping process, prompting a simultaneous rework of both.

What's new

  • Simplified onboarding: First-time installs now detect existing ChatGPT or Claude subscriptions, API keys, and local models, cutting configuration steps and letting users reach a first conversation faster.
  • Rebuilt browser app: The web interface opens directly into a conversation with the user's Claw, supporting live session follow-along, durable progress cards across reloads, and subagent activity tracking.
  • Shared cloud sessions (multiplayer): Teams can join live work, hand off tasks with context intact, and run sessions on paired devices or cloud workers while moving the workspace with them.
  • Interactive dashboards and widgets: Users can pin widgets to session dashboards, grant specific actions or network origins, and export rendered views as images.
  • Private credential requests: Agents can request credentials through masked prompts without exposing values in chat or model context, with an opt-in proxy limiting secret substitution to approved destinations.
  • Automation approvals: Recurring work can be approved once for an exact operation, with later inspection, revocation, and mandatory re-approval when the job changes.
  • Richer media handling: Audio and video stay attached across uploads, generated replies, playback, and reloads, with native controls on Apple and Android clients.
  • Memory and learning defaults: Personal conversation recall, grounded dreaming (background memory consolidation), and automatic self-learning are enabled by default on personal installs, with explicit disable controls.
  • CPU-scaled concurrency: Default top-level agent concurrency now sizes from available CPU parallelism, bounded between 8 and 16 simultaneous runs.
  • Explicit model allowlists: Per-agent policies and provider wildcards separate aliases and per-model settings from explicit modelPolicy.allow restrictions; openclaw doctor --fix migrates valid legacy restrictions without silently opening access when migration is incomplete.
  • Named agent setup: Users choose the first agent name during interactive onboarding or via --agent-name; legacy main-session history migrates to its clear owner.
  • Plugin trust and provenance: External plugin installs now show capabilities, source, version, and exact artifact before enablement; arbitrary executable sources require --force, while trusted ClawHub, bundled, official-catalog, and tracked-update sources skip the provenance warning but retain capability consent.
  • Session permission modes: Explicit modes keep restricted filesystem access anchored to the recorded workspace or worktree; new worktrees inherit configured policy unless a mode is selected.

Breaking changes and migrations

Two breaking migrations ship with 2.0. The OpenProse migration removes the bundled OpenProse plugin and /prose command; users run openclaw doctor --fix to clean stale configuration and follow the upstream Agent Skill migration, keeping existing .prose source files. The OpenAI route migration rewrites shipped codex/* and openai-codex/* model refs, provider config, stored sessions, and automation routes to openai/*, retaining Codex runtime intent and flagging conflicts for operator repair.

Upcoming deprecations target September 1, 2026: external plugins should migrate ahead of the SDK removal gate, with subpath imports moving to focused openclaw/plugin-sdk/ imports per the SDK migration guide.

Why it matters

OpenClaw 2.0 shifts the project from a rapid-cadence tool toward a platform that can serve both solo users and teams without losing context. The multiplayer session model — shared cloud sessions with workspace portability — addresses a gap the team felt internally while building the release: no way to bring another member into work without losing what the Claw already knew. That same architecture now lets families or teams collaborate on agents that span inboxes, calendars, messaging, and custom skills, all running on the user's own machines or chosen cloud workers.

Our take

The release's scale — half of all PRs ever merged — reflects a project that grew faster than its own release process could absorb. By pausing the cadence to harden the upgrade path for existing users, the team chose stability over velocity, a signal that OpenClaw is targeting production workloads. The real test will be whether the new plugin SDK gates and model allowlist migrations hold up across the ecosystem without fragmenting the community.

Sources