Horde Studio v12.6.0

Horde Studio 12.6.0

Version 12.6 replaces one-off text-provider integrations with a configurable provider layer, while correcting AWS Bedrock account routing and making model selection visible where providers are configured.

Custom providers

• Added a first-class Custom OpenAI-compatible text provider.
• Configure a friendly provider name, complete API base URL, bearer API key and optional JSON request headers.
• Discover models live from BASE_URL/models, search the returned catalog, or enter an exact model ID when a platform does not expose a catalog.
• Route Chat, Worlds, builders and Virtual Humans through BASE_URL/chat/completions using the same shared provider abstraction.
• Made the custom provider available as both the global text default and an independent Virtual Human conversation provider.
• Accept model catalogs returned as a plain array, { "data": [...] } or { "models": [...] }.

AWS Bedrock and NVIDIA NIM

• Added an optional complete Bedrock Mantle Base URL so account-specific endpoints no longer have to match Horde Studio's derived regional URL.
• Kept AWS Region as a fallback only when no explicit Mantle URL is supplied.
• Moved the searchable Default Text Model control beside the provider selector.
• Model browsing now follows the provider and credentials currently visible in Settings, even before saving.
• NVIDIA NIM and AWS Bedrock therefore expose their live model choices instead of appearing as key-only integrations.

Security and compatibility

• Require HTTPS for remote custom APIs while continuing to allow HTTP loopback servers.
• Validate additional headers as a bounded JSON object with string values.
• Store custom API keys and authentication headers through the protected credential path controlled by Remember Keys.
• Exclude custom credentials and authentication headers from exports and backups.
• Expanded the connection policy to support user-configured HTTPS providers without adding every host to Horde Studio itself.
• Added contextual help and Pip handbook coverage for custom APIs and account-specific Bedrock endpoints.

Validation

• JavaScript syntax checks pass.
• The complete Horde Studio audit suite passes.
• Custom-provider, cloud-provider, NanoGPT, contextual-help, security, World, Chat and Virtual Human regression coverage passes.
• The portable package includes all application, Labs, help, launcher and bridge files.

Horde Studio v12.5.0

Horde Studio 12.5.0

Version 12.5 makes Horde Studio’s optional local cognition genuinely useful, turns Pip into a dependable built-in guide, expands cloud-provider choice, and hardens the World and Virtual Human systems around the failures reported since 12.0.1.

Pip and contextual help

• Promoted Pip to a permanent main-sidebar workspace with a clearable conversation.
• Added a versioned, built-in Horde Studio handbook covering Chat, Virtual Humans, Worlds, Labs, providers, media, portability and troubleshooting.
• Product answers are grounded directly in verified local notes rather than rewritten by the 135M model.
• Added a second emergency core manual inside Pip’s runtime, so missing or stale handbook assets can no longer produce the old “ask Discord” dead end.
• Added contextual hover and keyboard-focus help throughout the application, including persistent explanations for Labs modes.

Horde Labs and the Embedded Tiny Brain

• Improved Embedded Tiny Brain installation persistence and launcher-based worker loading.
• Expanded safe micro-tier World assistance for actor-scoped movement, destination intent, travel, clothing and continuity signals.
• Added stricter structured-output cleanup, bounded JSON repair and deterministic fallbacks for malformed tiny-model output.
• Micro-model hints remain advisory: canonical World state is still validated by deterministic engine rules.
• Added clearer runtime, capability-tier and diagnostics reporting so expected skips are not presented as mysterious failures.

Providers

• Added NanoGPT text, image and neural-speech support with independent credentials and live model discovery.
• Added NVIDIA NIM as a first-class OpenAI-compatible text provider.
• Added Amazon Bedrock through its OpenAI-compatible Mantle endpoint with Bedrock API-key authentication and configurable AWS region.
• Text, image and speech selection remain independent. NVIDIA NIM or Bedrock text can be paired with another supported image provider.
• Added provider-specific connection tests, contextual guidance and safe credential persistence controls.

Worlds and immersion reliability

• Strengthened first-person movement resolution against the canonical location graph.
• Improved generic-room and local-destination matching so actions such as entering a bathroom do not jump to an unrelated named location.
• Prevented pronouns, quoted speech and another character’s movement from incorrectly moving the player.
• Expanded outfit replacement, addition and removal recognition while rejecting negated or merely discussed clothing changes.
• Improved streamed tool-call reconstruction, malformed argument repair, entity matching and schedule-aware arrival handling.

Virtual Humans

• Preserved independent conversation and photo providers across authored Humans and timelines.
• Improved time, silence, unanswered-message and ghosting awareness.
• Added stricter agency limits so multiple autonomous bubbles count as one unanswered turn and Humans do not message forever without acknowledgment.
• Retained situation-aware photo capture, model-specific image parameters, voice-format handling and media credit opt-outs.

Interface and release hardening

• Improved narrow-window layouts, World Studio navigation, message composers, focus states and contextual accessibility labels.
• Added explicit Play and Edit actions to Chat Library cards and retained the official Discord link in the main sidebar.
• Corrected the portable build manifest so Pip’s handbook and contextual-help runtime are included in every release ZIP.
• Simplified release automation so tag publication owns the release and portable asset without a competing branch workflow.

Validation

• JavaScript and Python syntax checks pass.
• Pip handbook, contextual help, cloud-provider, NanoGPT, Labs cognition, cross-mode UX, Virtual Human and World-intent regression audits pass.
• The actual localhost launcher serves Pip’s handbook and contextual-help files with HTTP 200 and the correct JavaScript content type.
• Browser verification confirms Pip answers “explain Horde Studio” from verified local knowledge.

Horde Studio v12.0.1

Horde Studio 12.0.1

Version 12.0.1 is a focused Embedded Tiny Brain and launcher reliability release. It fixes the browser-origin failure reported when Horde Studio was opened directly from index.html, repairs the official launcher’s Labs asset delivery, and adds regression coverage for both paths.

Embedded Tiny Brain

• Fixed the Failed to construct 'Worker' error caused by attempting to start the module worker from a file:// page with a null origin.
• Worker URLs are now resolved explicitly against Horde Studio’s local HTTP application origin.
• Direct-file mode now explains the actual browser restriction and directs the user to the included Start Horde Studio launcher.
• Removed the misleading implication that this origin error was necessarily caused by WebGPU, WASM, browser storage or connectivity.
• Failed worker construction now cleans up its pending request immediately instead of leaving stale work behind.

Launcher and portable app

• Added every Horde Labs runtime file and the Embedded Tiny Brain worker to the localhost bridge’s static application manifest.
• Confirmed the official launcher serves the module worker from http://127.0.0.1:43127 with the correct JavaScript content type.
• Updated application version metadata and cache-busting for 12.0.1.
• Rebuilt the portable package with the corrected launcher and Labs runtime.

Presentation

• Added the Horde Studio 12 launch poster to the README.
• Added an optimized Reddit-ready launch kit, derived from authentic application captures, with a hero poster, Living Worlds feature panel, Virtual Humans feature panel, Horde Labs feature panel and prepared launch copy.

Validation

• Labs cognition audit passes, including the direct-file recovery contract.
• MCP bridge audit passes with 15 checks, including coverage for all launcher-served Labs assets.
• JavaScript syntax and Python compilation checks pass.
• Browser verification confirmed the supported launcher path requests labs-embedded-worker.js successfully with HTTP 200.
• Portable archive contents and launchers verified before tagging.

Horde Studio v12.0.0

Horde Studio 12.0.0

Version 12 introduces Horde Labs: an optional, private local-cognition layer for Chat, Worlds and Virtual Humans. It also adds configurable status HUDs to ordinary chats, hardens provider image delivery, expands ComfyUI workflow support, and completes a release-wide accessibility and responsive-layout audit.

Horde Labs

• Added a shared Horde Labs control center without creating a separate or incompatible play mode.
• Connect a tiny local model through Ollama, LM Studio, llama.cpp, KoboldCpp or another OpenAI-compatible localhost server.
• Added Embedded Tiny Brain, with an in-app, opt-in download of SmolLM2 135M Instruct and WebGPU or CPU/WASM execution.
• Labs can assist Chat, Worlds and Virtual Humans independently in Off, Shadow, Assist or Audit modes where applicable.
• Small-model output is treated as an untrusted proposal: deterministic validators, evidence checks and bounded schemas remain authoritative.
• Added per-hour compute budgets, timeouts, graceful fallback and private diagnostics that never retain prompt text.
• Existing Horde Studio behavior remains unchanged when Labs is disabled or unavailable.

Optional HUDs for character chat

• Character Studio now includes a HUD & Meters editor for ordinary one-to-one chats.
• Authors can configure a title, status brief, up to eight bounded meters, colors and initial values.
• Each timeline maintains independent HUD state.
• Updates can be manual, driven by the main chat model, or proposed by Horde Labs.
• Grounding checks reject unsupported status changes rather than silently inventing state.
• The HUD collapses into an overlay on narrower screens and can be hidden during play.

Virtual Human improvements

• Added a per-timeline Text-message bursts option for choosing between natural consecutive bubbles and a single complete reply.
• Improved long-reply splitting so sentence boundaries are preserved without producing walls of text.
• Timeline controls expose clearer state and continue to isolate memories, relationships and lived history.
• Text, image and voice providers remain independently selectable.

Images, ComfyUI and local media

• Added named ComfyUI workflow profiles with create, duplicate, delete and quick-switch controls.
• The active profile preserves its workflow JSON and node mappings without overwriting other setups.
• Added a restricted localhost media proxy for provider-generated images that browsers cannot load directly.
• Remote media fetching is HTTPS-only, provider-allowlisted and protected against private or reserved network targets.
• Image responses continue to accept provider URLs, embedded data and base64 output while rejecting non-image task links.

Release-quality UI and accessibility

• Audited Chat Library, Worlds, Virtual Human, all three creator studios, live play, Photos & Voice, Labs and session setup at laptop viewport sizes.
• Fixed blank artwork placeholders by showing consistent initials in library cards, chat headers and message avatars.
• Named icon-only send, favorite, close and destructive controls for assistive technology.
• Correctly associated legacy visual form labels with their controls.
• Added consistent keyboard focus treatment and reduced-motion support.
• Fixed stale asset delivery with Version 12 cache-busting.
• Verified New Session Setup can be dismissed without committing a new life.
• Preserved toolbar wrapping and eliminated horizontal clipping across the audited flows.

Distribution

• Updated the portable builder to include every Horde Labs runtime and worker file.
• Version metadata, update checks, launchers and the local bridge now identify Horde Studio 12.0.0.
• The portable package remains local-first and requires no Node.js installation or build step.
• Embedded Tiny Brain downloads on demand and is cached by the browser; it is not bundled into the portable ZIP or backups.

Validation

• All JavaScript syntax checks pass.
• All authored-world, calibration, companion, engine, faction, living-world, movement, presentation, smart-input and Labs audit suites pass.
• Virtual Human audit: 159 checks pass.
• Core regression audit: 27 checks pass.
• Cross-mode UX audit: 9 checks pass.
• MCP bridge security and integration audit: 14 checks pass.
• Portable archive contents and platform launchers verified locally before tagging.

Horde Studio v11.5.0

Horde Studio 11.5.0

This release makes Horde Studio easier to navigate, makes visual Worlds more reliable, and gives Virtual Humans a genuine awareness of when you text, how long you disappear, and whether they are still waiting for you.

Virtual Humans that notice time

• Virtual Humans now receive exact message send, read and reply times in their own timezone.
• Late-night messages remain late-night messages: a text sent at 2 a.m. can be read the next morning with that context intact.
• Long gaps are surfaced naturally when a conversation resumes instead of being flattened into an ordinary next turn.
• Added deterministic silence stages for companions who are waiting, concerned, hurt or beginning to detach.
• Silence consequences evolve relationship state locally without adding paid model calls.
• Autonomous follow-ups stop after an unanswered outreach instead of endlessly messaging through silence.
• Multi-bubble replies count as one response turn, preventing false ghosting and follow-up spam.
• Added a per-timeline Silence has consequences control for players who want time awareness without relationship penalties.
• Added clearer delivery, read-time and waiting-state details throughout the Virtual Human interface.

World presentation reliability

• Fixed Enable visual presentation and generated visual settings failing to persist after saving or reopening World Studio.
• Expanded image-response decoding for dedicated image APIs, multimodal chat responses, nested content parts, inline data and base64 payloads.
• Fixed successful OpenRouter map-skin generations remaining stuck on Generating… with no displayed image.
• All world visual generators now reliably restore their controls after success or failure.
• Improved dialogue attribution using character aliases, grammatical focus, speech tags and carried conversational context.
• Ambiguous dialogue now remains ordinary prose instead of confidently assigning the wrong portrait.
• Cinematic narrative cards use translucent glass so authored room backgrounds remain visible.
• World play now displays the actual world name and artwork instead of a generic Dungeon Master identity.

Cleaner creator and library flows

• Added explicit Play and Edit actions to Chat Library cards and group rooms.
• Added consistent save-and-enter actions across Chat, Virtual Human and World studios.
• Repaired Character Studio alignment, responsive identity fields and styled media uploads.
• Improved World Studio’s basic-information layout and narrow-screen behavior.
• Added the official Horde Studio Discord link to the persistent sidebar.
• Preserved creator state when smart helpers normalize presentation settings.

Distribution and repository cleanup

• Added and verified the compact cross-platform portable build pipeline.
• Removed duplicate full-resolution world artwork while retaining optimized shipped assets.
• Removed generated archives and macOS metadata from the repository workspace.
• Repaired invalid Git metadata and compacted local repository storage.
• Restored the shared audit source harness to version control so regression suites work in fresh clones.
• The portable package remains self-contained and includes the local bridge and platform launchers.

Validation

• Virtual Human regression audit: 157/157 checks pass.
• Cross-mode creator UX audit: 7/7 checks pass.
• World presentation reliability audit: 6/6 checks pass.
• World visual-media audit: 12/12 checks pass.
• Smart-input UX audit: 11/11 checks pass.
• Contextual intent and tool reliability audit: 13/13 checks pass.
• Movement and location hierarchy audit: 36/36 checks pass.
• Portable archive contents, filenames and launch scripts verified locally before tagging.

Horde Studio v11.0.0

Horde Studio 11.0.0

This release turns Worlds into a portable visual sandbox while making cinematic conversations clearer, more personal and easier to author.

Visual worlds

• Added optional Classic and Cinematic world presentation modes.
• Added portable per-location backgrounds, NPC portraits and map skins.
• Added upload, generation, replacement and cleanup tools for world media.
• Embedded media now survives world exports, imports, full backups and restored saves.
• Added searchable provider-aware world image-model selection, with Nano Banana 2 Lite as the recommended default.
• Added responsive cinematic layouts that preserve the existing gameplay HUD and semantic map.

Dialogue and personas

• Added portrait dialogue cards for recognized NPC speech.
• Added a configurable dialogue color to every NPC.
• Added persona colors and player dialogue cards using the selected persona name and avatar.
• Replaced fragile quote matching with discourse-aware speaker detection across paragraphs and multiple quotes.
• Added explicit screenplay-style Name: dialogue support.
• Retired the unfinished Visual Novel option; legacy saves automatically fall back to Cinematic.

Policy Panic at Bramble & Pike

• Added a complete small-town workplace-comedy sandbox as an included showcase world.
• Includes 23 locations, 8 scheduled autonomous NPCs, 7 starting lives, 6 factions and 15 lore entries.
• Includes 13 embedded cinematic assets covering the office, major rooms and the full principal cast.
• Added migration logic that upgrades the bundled world without overwriting creator-customized NPC colors.

Creator experience

• Replaced major free-form model fields with searchable provider catalogs while preserving exact custom model IDs.
• Added smarter controls for schedule days, location metadata, starting lives, stat overrides and colors.
• Added dedicated visual-media and smart-input regression audits.
• Added normalized image uploads and optimized generated media to keep portable worlds practical.

Updates and distribution

• Added a cached, non-blocking GitHub Release update check.
• New-version notices provide View update and Later actions and remain silent offline.
• Added local bridge routes for the included world, diagnostics, favicon and portable world download.
• Added an automated tag-based GitHub Release workflow.

Validation

• JavaScript and Python syntax checks pass.
• World visual-media audit: 12/12 checks pass.
• Smart-input UX audit: 11/11 checks pass.
• Policy Panic v4 bundle integrity, embedded media and dialogue colors validated.