Agent Zero Release Notes

Remote Control & Connectivity

Renamed Remote Link to Remote Control with a refactored tunnel provider system. Cloudflare Tunnel, Microsoft Dev Tunnels, Serveo, and Tailscale are available through the Remote Control selector.
Advertise Agent Zero version in connector handshakes so CLI clients can detect version mismatches with the connected server.
Expose installed plugin toggles via a protected API endpoint that lists installed plugins and supports toggling, while refusing changes to protected plugins like the connector.
Expose chat tab metadata in the connector API — returns chat number, honest chat name, and project metadata for terminal clients.

Extracted document_query into a standalone _document_query plugin with a parser strategy pattern, MIME-type registry with priority routing, five parser implementations (PDF, HTML, Text, Image, Unstructured), and all config externalized.
Added LiteParse as the preferred parser with legacy fallbacks, subprocess isolation to protect the Web UI from native parser crashes, bounded concurrency, and adaptive OCR heuristics that disable OCR for large text-rich PDFs.
Progressive skill loading — moved long tool instructions into a document-query skill with a compact tool prompt stub.
Expanded settings panel exposing parser, retrieval, fetch, LiteParse/OCR, and fallback controls in the plugin UI, plus a generated thumbnail.
Bounded long PDF processing — automatically disables LiteParse OCR above a configurable page threshold and uses adaptive index chunk sizing for large documents.
Pinned LiteParse to 2.0.3 and centralized dependency management in root requirements (removed plugin-level requirements file).

Durable screenshot artifacts — browser, desktop, computer-use, and vision screenshots are now materialized into chat-scoped artifacts so historical image references survive temporary screenshot pruning.
Background computer-use tool contract — exposes native window listing, indexed window state, and element_action dispatch so agents prefer structural targeting over screenshots.
Hardened internal Xpra desktop control with structured-first workflows, batched desktopctl sequences, and normalized resize requests to prevent narrow portrait sizes from distorting the X11 display.
Preserved computer-use capability metadata — stores and returns contract version and capabilities, teaching the remote computer-use prompt to prefer structured dispatch.
Fixed Xpra canvas sizing from backend-normalized display dimensions to prevent stretched desktop views.

Improved coding-agent verification discipline — added prompt guidance for reading specs/tests first, making narrow edits, verifying exact artifacts, cleaning temporary work, and recovering from tool failures.
Rebalanced coding discipline into the core prompt so all Agent Zero instances carry verification and final-report habits by default.
Raised code execution timeouts for builds, installs, servers, tests, and training jobs. Long-running work is steered toward output polling.
Aligned remote exec timeouts with code execution timeout groups for connected CLI sessions.

Improved auto-memory extraction quality — filters passive fragment memories, action history, temporary artifacts, and local runtime paths before consolidation. Tightened prompts toward durable preferences, stable project facts, and reusable solutions.
Added timestamp-based staleness guidance — treats memory timestamps as soft recency signals, preferring newer facts for mutable conflicts while preserving old durable preferences.
More explicit memory_forget usage guidance to improve memory consolidation around stale entries.

Unified OAuth account management with a shared provider-registry-backed status summary, redesigned settings modal with provider rows and account-backed model slots.
Provider-aware OAuth accounts in onboarding — supports provider-specific setup fields, device-code and browser callback flows, and model loading by provider.
Added extensible OAuth providers framework. Bring your xAI account, Google Cloud project, or GitHub Copilot account.

Simplified skills selector layout — replaced Visible/Pinned mode switch with a single continuous view; pinned skills appear first, with hide/show via an inline eye control.
Made Office, Desktop, and Editor plugins toggleable by removing forced always_enabled from their manifests.
Include top-level AGENTS.md in project system instructions with a UI checkbox and backwards-compatible defaults.

Clean desktop SSH agent state during self-update — removes stale runtime entries best-effort so missing or unexpected paths don't block updates.
Clean legacy Desktop agent sockets during self-update — covers current and retired Desktop profile state plus legacy Office-owned layouts, preserving GnuPG key material.
Added AGENTS.md DOX contracts across the repository to formalize ownership and local work contracts; migrated legacy deep-dive docs to the new layout.
Moved model config snapshot sync into the store for cleaner UI-only field initialization and dirty tracking.

Fixed per-subagent model configs.
Fixed document_query prompt spelling — corrected optmimize → optimize and cleaned legacy test references.

Reverted builtin ACP plugin — the Agent Client Protocol stdio plugin was added and subsequently reverted in this release cycle.