ZERO ONE 7.9.2

ZERO ONE 7.9.2

• Adds Browser Pilot directly inside ZERO ONE. It controls only the isolated tab the user grants, works through OpenZero's dedicated browser planner, and never needs a silently installed browser extension.
• Sends compact page structure and labels to the planner without form values; blocks password, payment, secret, file and CAPTCHA controls.
• Pauses cross-site navigation, personal-data entry and consequential actions for explicit one-time approval, includes an in-page STOP control, and ends every run at a 12-step safety limit.
• Keeps the optional Chrome/Brave OpenZero Tab Pilot as a separate user-approved extension for external-browser workflows.
• Upgrades the visible Settings update control into a verified one-click path. ZERO ONE accepts only an exact official stable-release package whose GitHub asset digest, SHA256SUMS.txt, declared size and downloaded bytes all agree.
• Starts the verified Windows one-click installer only after confirmation and preserves existing settings, isolated workspace sessions and model data across the upgrade.
• Preserves model separation: OpenZero Gemma4 E2B remains the recommended lightweight local Assistant, OpenZero Ministral 8B remains the full OpenZero server default, and the quality-rejected Fusion model stays blocked.
• Adds unit and source-contract coverage for browser action boundaries, secret-field exclusions, tab partition isolation, blocked downloads and verified update supply-chain checks.

Packages remain unsigned public previews until Authenticode, Apple notarization and Linux repository signing are available. Windows SmartScreen or macOS Gatekeeper may warn. Download only from the authenticated GitHub release; the in-app updater verifies integrity but does not create a publisher signature.

Browser Pilot is governed browser assistance, not unrestricted autonomy. It cannot bypass authentication, CAPTCHA, payment, secret-entry or operating-system permissions, and it does not silently submit or publish consequential changes.

ZERO ONE 7.9.1

ZERO ONE 7.9.1

• Makes ZMail a first-class workspace: opening Assistant mounts ZMail so inbox actions are available without losing the current view, and the ZMail Home control returns to the mailbox cleanly.
• Keeps the one-time zSign sign-on handoff inside the isolated ZMail session partition and allows only the reviewed zSign origin through the desktop boundary.
• Restricts saved-login injection to ordinary sign-in forms, excluding password-reset, password-change and multi-password forms.
• Limits inbox inspection to the messages already rendered in the ZMail workspace; it does not add background mailbox scraping or automatic sending.
• Fixes the explicit resource-safety action so it unloads every running published OpenZero model plus a currently selected custom local model, while leaving unrelated Ollama workloads alone.
• Preserves model separation: OpenZero Gemma4 E2B Agentic Q4_K_M remains the recommended lightweight local Assistant, and OpenZero Ministral 8B Runtime Agent remains the separate full OpenZero default. The selector blocks the quality-rejected Fusion model and Qwen3 1.7B release.
• Adds regression coverage for the ZMail/zSign partition boundary, login-form exclusions, workspace actions and all-model unload behavior.

Packages remain unsigned public previews until Authenticode, Apple notarization and Linux repository signing are available. Windows SmartScreen or macOS Gatekeeper may warn. Download only from this authenticated GitHub release and verify the package against SHA256SUMS.txt.

Direct local Ollama chat generates text on this computer. It cannot independently use OpenZero tools. Full OpenZero, zSign and Tab Pilot remain separately permission-gated surfaces.

ZERO ONE 7.9.0

ZERO ONE 7.9.0

• Keeps OpenZero Gemma4 E2B Agentic Q4_K_M as the recommended lightweight local Assistant and OpenZero Ministral 8B Runtime Agent as the standard full OpenZero server model.
• Separates the local Assistant selection from the full OpenZero server model, so connecting the full runtime cannot silently replace everyday local chat.
• Makes explicit Local mode authoritative even when a server credential is also stored, preserving an operator-selected custom Ollama model without silently routing its prompt to the server.
• Preserves legacy OpenZero server routing and its selected server model when upgrading settings that predate the explicit local/server mode field.
• The rejected Fusion model and Qwen3 1.7B release are blocked at the local-model boundary and excluded from the visible selector after response-quality testing.
• Adds low-memory, balanced and performance profiles, reports relevant loaded Ollama models, unloads competing OpenZero models before chat without disturbing unrelated Ollama workloads, prevents concurrent local chats and exposes an explicit unload action.
• Updates first-party copy and packaged smoke tests to validate the actual Gemma E2B default instead of the historical Qwen default.
• Improves keyboard focus visibility for links, selects and disclosure controls, and makes the always-visible version label easier to read.
• Keeps historical release evidence explicitly separate from claims about this source revision. The 7.9.0 build remains an unsigned public preview until the documented signing and final-candidate gates are complete.

Direct local Ollama chat generates text on this computer. It does not independently gain OpenZero tools, browser control or autonomous action authority. Full OpenZero and Tab Pilot capabilities remain separately configured and permission-gated.

ZERO ONE 7.8.9

ZERO ONE 7.8.9

• Uses OpenZero Gemma4 E2B Agentic Q4_K_M as the recommended lightweight ZERO ONE Assistant model.
• Keeps OpenZero Ministral 8B Runtime Agent as the full OpenZero server/operator default.
• Adds explicit local Assistant choices for Gemma E2B, Ministral 8B, experimental Qwen3 1.7B, and legacy Gemma E4B. The known-bad Fusion model is deliberately excluded.
• Answers simple greetings instantly without loading a multi-gigabyte model.
• Replaces stale Gemma E4B download and timeout guidance with model-aware wording.
• Migrates the managed 7.8.8 Ministral Assistant selection to Gemma E2B when E2B is already installed, while preserving explicit custom choices.

Local comparison on the target CPU found Gemma E2B materially safer for quick chat than the Qwen3 1.7B fine-tune: it correctly explained GGUF, honestly refused live-web access, and did not disclose the supplied system prompt. Qwen remains available only as an experimental manual choice.

ZERO ONE 7.8.8

ZERO ONE 7.8.8

• Makes OpenZero Ministral 3 8B Runtime Agent Q5_K_M the default local Assistant for new installs.
• Safely migrates recognized older managed defaults, including OpenZero Gemma4 E4B, when the new model is already installed; explicitly selected custom models remain untouched.
• Keeps OpenZero Gemma4 E4B available as a compatibility fallback.
• Aligns the embedded OpenZero panel and ZERO ONE quick chat on the same recommended model.
• Preserves the always-visible version label and review-first update center introduced in 7.8.7.

The Ministral release uses unchanged upstream weights plus an embedded OpenZero runtime template. It is a runtime-template edition, not a weight fine-tune. Full OpenZero tool execution remains permission-gated by the OpenZero runtime; the quick local Assistant does not independently execute tools.

ZERO ONE 7.8.7

ZERO ONE 7.8.7

• Shows the installed version persistently in the desktop sidebar.
• Adds a clear Version & Updates section with manual status checking.
• Distinguishes up-to-date, update-available, and temporarily-unavailable states.
• Keeps updates review-first: ZERO ONE never downloads or installs an update silently.
• Preserves existing settings, local models, encrypted credentials, and workspace sessions during an in-place upgrade.

ZERO ONE 7.8.6

ZERO ONE 7.8.6

• Adds OpenZero Ministral 8B Runtime Agent Q5_K_M as an explicit, optional local Assistant selection.
• Keeps the behavior-tested Gemma4 E4B local Assistant as the default; the 8B runtime edition is for capable CPUs and is not silently selected.
• Preserves an intentional published OpenZero GGUF selection at startup instead of replacing it with Gemma.
• Uses the selected model for the in-app download action and reports the selected local model during setup.
• The Ministral release is a runtime-template edition: upstream model weights are unchanged. Local chat remains separate from the permission-gated full OpenZero orchestration runtime.

ZERO ONE 7.8.5

ZERO ONE 7.8.5

• Adds the operator's sovereign research ethics to the local assistant contract: privacy, minimum disclosure, evidence/provenance, uncertainty, and truthful completion status.
• Explicitly prevents the local assistant from requesting, repeating, inferring, or exposing passwords, tokens, private keys, or hidden credentials.
• Keeps privileged autonomous and secret-backed operations behind the permission-gated OpenZero runtime instead of exposing them to raw local chat.
• Retains capability-honesty, policy-leak, role-injection, repetition, and bounded CPU-response protections.
• Preserves existing ZERO ONE settings and encrypted workspace-login storage across installation.

ZERO ONE 7.8.4

ZERO ONE 7.8.4

• Makes local capability boundaries explicit: no browsing, live databases, files, terminal, tools, or external actions through quick chat.
• Prevents false claims of web or filesystem access in identity and capability answers.
• Bounds local replies to 256 generated tokens for practical CPU latency.
• Raises the local request timeout from 45 to 120 seconds so correct CPU responses are not reported as failures.
• Passed an 11-case behavioral suite covering identity, capability honesty, factual answers, exact formatting, policy secrecy, repetition, latency, and multi-turn memory.
• Uses deterministic, truthful local answers for identity/capability questions and requests for hidden instructions, preventing model-generated capability hallucinations or copied policy text.

ZERO ONE 7.8.3

ZERO ONE 7.8.3

• Adds a dedicated conversational system prompt for local Ollama models.
• Removes leaked OpenZero operator-policy responses from chat history and rejects them as empty model output.
• Increases the local reply budget and adds repetition control for models that spend tokens on internal reasoning.
• Recognises both legacy Zero-* and current OpenZero-* published GGUF repositories as local models when installed.
• Uses the locally benchmarked OpenZero Gemma4 E4B release as the reliable default; the smaller experimental fine-tunes remain optional pending stronger conversational evaluation.