Product changelog

Every update moves ideascloser to working hardware.

A record of IterXAI desktop changes formally released from main, focused only on completed capabilities, experience, and reliability.

Latest stable releasev0.1.99September 8, 2026

September 2026

Project design, compilation and conversation reliability

Improved the workflow from component documentation and project connections to compilation and simulation, with activation-code redemption and clearer feedback and recovery.

  • Redeem activation codes for bonus tokens, view bonus balances separately, and enforce one redemption per account and code.
  • Improve conversation reconnection and tool calls, reduce duplicate execution, and distinguish rate limits from connection failures.
  • Improve component intake, project snapshots and functional pin mapping while isolating unsaved intake when switching projects.
  • Improve project connections, board component placement and network editing while preserving confirmed positions and electrical constraints.
  • Improve reuse of compilation components and simulation setup, including long paths, debugging recovery and board startup fixes.
  • Add virtual Bluetooth controller interactions and improve feedback submission, discussion and progress visibility.

Stable 0.1.98: experience and reliability updates

This stable release from main consolidates 0 completed changes across experience and reliability.

  • Improved installation, saving, interface feedback, and failure recovery.

Module simulation, safer updates, and a unified status entry

Expanded managed firmware simulation and the production module runtime, tightened installer, updater, asset, and AI verification boundaries, and unified every service-status entry under the IterxAI website.

  • Expanded managed firmware simulation across more controller families, including controller-only projects and additional device profiles.
  • Shipped the production module runtime with platform-specific manifest and hash verification for compile, upload, and simulation assets.
  • Added disk-space, installation-path, and stale-process protection to reduce incomplete or failed in-place updates.
  • Simulation and AI verification now require the real project board identity and an explicit terminal state, reducing false success and duplicate work.
  • Dual-platform releases now remain bound to an exact source revision with staged failure handling.
  • Every service-status action now opens iterx.ai/status instead of maintaining a separate legacy address.

Managed simulation recovery restores board validation

The stable app can prepare its managed simulation runtime again, allowing board projects to continue validation without relying on development-only files.

  • Stable packages can download and verify the approved simulation runtime when it is missing.
  • Simulation runtime installation stays isolated from programming environments, project files, and unrelated tools.
  • Existing installations are reused, while failures keep a clear recovery path.

Single-board simulation, environment tools, and settings fixes

Fixed projects containing only a controller being rejected for missing connections, enabled direct environment inspection and installation from the Agent, and prevented delayed focus from interrupting settings input.

  • Controller-only projects without external wiring can now enter firmware simulation without unrelated power, ground, or net checks.
  • Simulation startup failures now distinguish project configuration problems from local runtime problems.
  • The Agent can directly inspect and install programming environments instead of stopping at a generic warning.
  • AI model settings only apply automatic focus while the form is idle, preserving the field the user is editing.

Stable changelog publishing restored

Fixed a syntax error in the stable-release metadata script so published installers can be accompanied by complete, traceable changelog entries in both languages.

  • Corrected changelog metadata handoff in the release flow to prevent script failures.
  • Added regression coverage for stable version, channel, and publication parameters.
  • Kept changelog entries tied to the matching installer version and source revision.

August 2026

Release reliability and extension isolation are stronger

This release reduces transient publication failures and prevents one malformed extension from disrupting the full AI workflow.

  • Short-lived timing service failures are retried during publication.
  • Malformed or incompatible extension content is isolated.
  • Extension parsing failures no longer interrupt normal conversations and project work.

Project truth, per-file drafts, and atomic commits are closed end to end

Project membership, electrical networks, simulation, and board evidence now share one truth, while AI changes can be reviewed per file before one atomic commit.

  • Project members, electrical netlists, simulation, and board output use one verifiable truth.
  • Delivery targets bind project checks and truth snapshots to reduce cross-stage drift.
  • AI supports per-file drafts followed by one atomic commit after approval.
  • Tools, models, and domain guidance narrow dynamically for each task.
  • Model-gateway failures are layered so configuration, quota, and upstream causes are easier to distinguish.
  • Native tool workspaces and text-format compatibility are more reliable.

Dynamic tools and long-task recovery are more reliable

AI tools are no longer blocked by unrelated inventory state, while task recovery, model switching, and programming diagnostics provide clearer feedback.

  • Dynamic tools follow actual execution readiness instead of an unrelated inventory.
  • Long tasks recover more safely after interruption.
  • Model-switching copy and repeated selections are handled more clearly.
  • Programming diagnostics report more accurate runtime state.
  • Component-card metadata no longer clips in narrow layouts.

Consent, runtime identity, and failure classification are consolidated

Sign-in consent, programming-environment identity, AI compatibility, and failure categories now follow clearer product boundaries.

  • First use includes agreement and privacy confirmation.
  • Programming-environment identity is restored and remote availability uses one publication truth.
  • Self-configured model access follows one response protocol.
  • The old conversation path is retired and failures use consistent categories.
  • Simulation button input and controlled forms behave more reliably.

Installation and model selection are stable again

Fresh-install path checks, uninstall behavior, build dependencies, and the model menu were aligned in one reliability pass.

  • Fresh installation no longer mistakes an ancestor folder for an existing app directory.
  • Uninstall can build and clean shared runtime state correctly.
  • Release builds use the fixed runtime dependency contract again.
  • Model selection keeps a clear text-first entry and loads faster.
  • Background correction no longer blocks users from starting work.

Component intake, tool recovery, and reload saving are more reliable

Key interruption points are now protected, from component evidence intake to tool recovery, simulation navigation, and app reloads.

  • Component intake follows submit, recognition review, and publish stages with a tighter merge boundary.
  • Missing facts and evidence sources are called out before they reach a project.
  • Transient tool-delivery failures can recover without ending the whole task.
  • Simulation runtime no longer depends on staying on one page.
  • A close-and-save handshake runs before reload, reducing unsaved work loss.

Programming, simulation, and pin editing form a complete workflow

After several main-branch refinements, managed programming, isolated simulation, board networks, and AI execution state are more predictable.

  • Common programming environments support real builds, offline reuse, uninstall confirmation, and project isolation.
  • Simulation runs are isolated, with improved input behavior, visual state, and canvas stability.
  • Board editing adds network changes, multi-row pin spacing, function mapping, and evidence intake.
  • Programming code no longer gets overwritten or lost on exit.
  • AI progress, failure diagnosis, model selection, and low-risk notices are clearer.

AI collaboration and the hardware workflow receive a broad upgrade

AI conversations, goal management, programming setup, simulation performance, and PCB editing move further into one continuous workflow.

  • AI supports parallel approvals, native collaboration, long-task tools, and stronger project context continuity.
  • Goal and plan input are unified, while active goals can be managed independently.
  • Programming configuration and font scaling arrive, and simulation routing moves off the main interface thread.
  • Package editing adds pad dragging, crosshair positioning, network highlighting, and steadier saving.
  • Fabrication output, project creation, sign-in recovery, and page lifecycle handling are more reliable.

Stable AI metadata and protocol compatibility are fixed

Model information and bundled protocol snapshots are consistent again, avoiding parsing failures caused by text-format differences.

  • Stable packages load company model metadata correctly.
  • Bundled protocol snapshots use a consistent line-ending format.
  • Model selection and capability descriptions remain available after launch.

AI Agent, simulation, and board workflows receive a major upgrade

Native AI execution, project attachments, programming environments, simulation interaction, and fabrication now form a more continuous desktop experience.

  • The AI Agent supports native conversations, project tools, model selection, long tasks, and recovery from output limits.
  • Project attachments can move through strict review and finalization.
  • Environment setup, permission recovery, real builds, and offline reuse are more reliable.
  • Simulation adds bounded external inputs, a data sidebar, runtime feedback, and steadier component behavior.
  • Board routing, pad functions, fabrication output, and cross-stage project facts stay aligned.
  • Compact editors, resizable input, language copy, and waiting feedback are improved throughout the app.

Solution quantities become traceable component instances

Multiple units of one component now become distinct instances that remain aligned across simulation, board editing, and filtering.

  • Solution quantities expand into separate component instances.
  • Simulation pins and solution membership share the same facts.
  • Board editing can filter individual instances.
  • Conversations show immediate waiting feedback after a message is sent.

July 2026

Component identity and simulation visuals stay aligned

Simulation now uses the same project component identity and visual evidence, reducing cross-stage mismatches.

  • Solution, simulation, and board stages share one component identity.
  • Visual evidence is selected from component facts instead of ambiguous names.
  • Component relationships stay correct after stage changes.

Component projection from solution to board is more complete

Solution components now carry into the board view with clearer visuals, names, and editable connection state.

  • The board view synchronizes components from the active solution.
  • Component images retain their proportions with a reliable visual fallback.
  • Stale empty netlists are rebuilt from the current solution.
  • Conversation boundaries and stage controls are easier to understand.
  • Generic controller naming stays consistent and avoids misleading model assumptions.

Safe file intake, screenshot feedback, and live drafts arrive

Projects can safely receive external files, fabrication edits retain drafts, and feedback can include visual evidence.

  • External files can be dragged safely into the active project.
  • A default workspace mode can be selected after registration.
  • Physical edits and simulation connections are preserved continuously.
  • Feedback can include screenshot attachments.
  • Project navigation and programming-environment writes are better isolated.

Permissions, netlist saving, and accessibility are improved

The stable build tightens local resource access while preserving board data and keyboard focus across navigation.

  • Programming environments use consistent current-user permissions.
  • Board editing preserves the explicit netlist instead of overwriting it.
  • Local resources are read only within the active project.
  • Hidden stages no longer retain focus, improving keyboard navigation.
  • Runtime font rendering no longer depends on an external request.

Programming recovery, physical pins, and fabrication checks are hardened

Several interruption-prone paths now behave consistently, from environment recovery and device coordination to pin calibration and fabrication checks.

  • Device discovery pauses during programming jobs to avoid contention.
  • Recovery preserves the selected target and sealed environment without affecting other projects.
  • Physical pins, module geometry, and network relationships now reach board-level checks.
  • Project creation, account callbacks, and failures provide clearer recovery actions.
  • File writes are serialized to reduce concurrent save conflicts.

Automatic update compatibility is improved

Update packages now exclude unnecessary system metadata, reducing failures during extraction and replacement.

  • Update payloads retain only files required by the app.
  • Hidden metadata no longer interferes with package verification.
  • In-place updates complete more reliably.

Pin projection and routing are more reliable

Component pins stay consistent in the board view, with steadier handoff between routing and later edits.

  • Pin projection no longer shifts or disappears unexpectedly.
  • Network relationships stay consistent before and after routing.
  • Board editing avoids unnecessary recomputation and state drift.

Build and data-writing checks are more precise

Code compilation and data-partition capability are now evaluated separately, producing more accurate failures.

  • Normal builds are no longer blocked by unrelated data-writing checks.
  • Data-partition capability is verified only when needed.
  • Failures point to a more accurate recovery step.

Global appearance controls arrive with safer simulation saving

Desktop appearance can be adjusted consistently, while simulation autosave no longer interrupts the current task.

  • Custom themes and global appearance controls are now available.
  • Layout autosave no longer triggers cross-stage navigation.
  • The current simulation context stays stable while saving.

Themes, project recovery, and fabrication checks receive a broad upgrade

This release improves long-session comfort, conversation recovery, firmware data writing, component evidence, and board readiness.

  • Themes now cover primary pages, dialogs, sidebars, and status feedback consistently.
  • Conversations stop, continue, stay isolated, and recover more reliably.
  • Firmware data partitions can be generated and written.
  • Simulation settings stay with each component instance, while fabrication checks block incomplete plans.
  • Project renaming, visual programming retention, and sign-in recovery are more reliable.

Only stable releases from main appear here. Development and prerelease work is never shown early. Older installers may no longer be available from the current download entry.