Version 0.4.0 reworks the built-in local AI engine: instead of shipping one fixed model inside the app, you now download and manage local models on demand from Settings. - The app no longer bundles a local AI model, keeping the download smaller. Download the model you want, when you want it. - Built-in model management: browse a catalog of local models in Settings ▸ Backend ▸ Built-in (Local), download with a live progress bar, choose which installed model is active, and delete models you no longer need. - Three models are offered to start — Qwen3.5 4B, Qwen3.5 9B, and Qwen2.5 Coder 7B Instruct (Q4_K_M GGUF) — downloaded directly from Hugging Face. - Downloads are verified against a SHA-256 checksum, so a corrupted or interrupted download won't be used. - Downloaded models are stored in the app's container and are kept across app upgrades — updating PrivyAPIStudio does not re-download models you already have. - The model catalog is defined in a configuration file rather than hardcoded, making it straightforward to adjust the offered models in future updates. - Existing users who relied on the previously bundled model are prompted to download a model on first launch; the built-in provider is marked as needing a model until one is installed. - Internal dependencies are now version-pinned to remote releases (including the new model-management component), improving build reproducibility. - Kept the existing macOS 13 minimum system requirement.

Version 0.4.0 reworks the built-in local AI engine: instead of shipping one fixed model inside the app, you now download and manage local models on demand from Settings. - The app no longer bundles a local AI model, keeping the download smaller. Download the model you want, when you want it. - Built-in model management: browse a catalog of local models in Settings ▸ Backend ▸ Built-in (Local), download with a live progress bar, choose which installed model is active, and delete models you no longer need. - Three models are offered to start — Qwen3.5 4B, Qwen3.5 9B, and Qwen2.5 Coder 7B Instruct (Q4_K_M GGUF) — downloaded directly from Hugging Face. - Downloads are verified against a SHA-256 checksum, so a corrupted or interrupted download won't be used. - Downloaded models are stored in the app's container and are kept across app upgrades — updating PrivyAPIStudio does not re-download models you already have. - The model catalog is defined in a configuration file rather than hardcoded, making it straightforward to adjust the offered models in future updates. - Existing users who relied on the previously bundled model are prompted to download a model on first launch; the built-in provider is marked as needing a model until one is installed. - Internal dependencies are now version-pinned to remote releases (including the new model-management component), improving build reproducibility. - Kept the existing macOS 13 minimum system requirement.

Version 0.3.1 is a privacy and correctness release that strengthens credential handling, fixes request building, and tidies the codebase. - API request credentials (bearer/JWT tokens, basic passwords, API keys) saved with an endpoint are now stored in the macOS Keychain instead of the local database. Existing saved credentials are migrated automatically on first launch. - Request history no longer records secrets in plaintext. API keys placed in the query string, and secret values appearing in request or response bodies, are masked before history is saved. - Query parameters and request headers now preserve repeated names, so requests like `?id=1&id=2` or multiple headers with the same name are sent exactly as entered. - Fixed URL building so a query string already present in the path is kept when additional parameters are added. - Continued sending sensitive request headers (Authorization, Cookie, API keys, and similar) redacted in history, now alongside the broader URL and body masking. - Internal refactor that splits the largest source files into focused modules, improving maintainability with no change to behavior. - Kept the existing macOS 13 minimum system requirement.

Version 0.3.0 focuses on making AI-assisted API work easier to read, easier to control, and less intrusive when you need more workspace. - AI answers now render as compact GitHub-flavored Markdown, with clearer headings, lists, links, blockquotes, inline code, and fenced code blocks. - Added an optional Thinking mode for questions that need deeper analysis. The setting is remembered between sessions. - Thinking mode presents a short, collapsible reasoning summary separately from the final answer, keeping responses structured and easy to scan. - The AI composer now shows live `Thinking…` or `Answering…` progress directly inside the input area. - Improved AI response instructions so answers lead with the recommendation, preserve exact HTTP methods and API paths, and use properly labeled code blocks. - The right-side Assistant, Diff, and History panel can now be collapsed from the toolbar and restored when more context is needed. - Right-panel visibility is remembered per window, and the panel remains resizable when open. - Response Body and Headers now share the same compact copy control. The button appears only when the pointer is over the response content and stays hidden for empty responses. - Upgraded the local AI stack to ProviderKit 0.6.1, LlamaProviderKit 0.1.1, and LlamaKit 0.5.1 using pinned remote release tags. - Moved local model integration to the dedicated LlamaProviderKit package, keeping the core provider layer smaller and more modular. - Kept the existing macOS 13 minimum system requirement.

Version 0.2.1 is a polish release focused on privacy, speed, and reliability for everyday use. - API keys are now stored securely in the macOS Keychain instead of plain settings. - Saved request history automatically hides sensitive values such as tokens, passwords, and API keys, so you can share or screenshot a request safely. - The built-in AI assistant responds faster: the local model now stays ready between questions instead of reloading each time. - Importing large API specifications no longer freezes the window — the app stays responsive while it works. - Re-importing a specification now keeps any endpoints you created by hand and only refreshes the imported ones. - Requests to slow or unresponsive servers now time out cleanly instead of hanging indefinitely. - Non-text (binary) responses, such as images, now show a clear summary instead of appearing blank. - Request history is trimmed automatically over time, so the app stays lean. - The app now starts up more reliably, with a clear message if local storage is unavailable instead of failing silently. - New "View Licenses" entry in Settings lists every bundled open-source library and AI model, each with its full license text.

PrivyAPIStudio 0.2.0 focuses on making local AI assistance work the moment the app is opened, while keeping Ollama and LM Studio fully supported. Added a new Built-in (Local) backend powered by LlamaKit (llama.cpp) with Metal acceleration on Apple Silicon. Bundled the Qwen3.5-0.8B Q4_K_M GGUF model directly in the app, so AI features work offline without any external server. Set the Built-in backend as the default on fresh installs; new users no longer need to install Ollama or LM Studio before using AI assistance. Added a forward migration for existing users: the Built-in backend now appears automatically in Settings without changing the user's currently active provider. Redesigned the Settings → Backend panel with a clearer backend picker, per-backend description text, and separate field layouts for built-in vs. network providers. The Built-in backend shows a read-only bundled-model card with a Metal acceleration indicator instead of an editable Base URL / Model row. The "Refresh Models" button is now simply "Refresh" and works for all three backends, including the built-in engine. Updated AI configuration validation so the Built-in backend is considered valid as long as a bundled model is present (no Base URL required). Fixed local model resolution so GGUF filenames containing dots (such as Qwen3.5-0.8B.Q4_K_M) are matched correctly inside the app bundle. Cleaned up the macOS project layout: removed the duplicate Package.swift / Package.resolved next to the .xcodeproj, which was causing Xcode to report "Missing package product" for ProviderKit, GRDB, MarkdownUI, and Yams in the IDE. Moved the local ProviderKit dependency into XcodeGen's localPackages section and added a scripts/fix-pbxproj.sh post-processor to keep local Swift package references stable across regenerations.

Added a dedicated Auth tab to the request editor. Added support for Basic Auth, Bearer Token, JWT Token, API Key, and No Auth request modes. API keys can be sent either as headers or query parameters. Authentication values support existing {{variable}} interpolation, so tokens and keys can come from global, project, or environment variables. OpenAPI imports now infer request authentication from securitySchemes and Swagger securityDefinitions when possible. Saved API records now persist authentication configuration alongside params, headers, and body data. Request execution now generates authentication headers and query values at send time, keeping manual headers and explicit auth settings separate. Added tests for authentication request preparation, OpenAPI security inference, and auth persistence.

PrivyAPIStudio 0.1.5 focuses on making authenticated API testing easier and less error-prone. - Added a dedicated Auth tab to the request editor. - Added support for Basic Auth, Bearer Token, JWT Token, API Key, and No Auth request modes. - API keys can be sent either as headers or query parameters. - Authentication values support existing `{{variable}}` interpolation, so tokens and keys can come from global, project, or environment variables. - OpenAPI imports now infer request authentication from `securitySchemes` and Swagger `securityDefinitions` when possible. - Saved API records now persist authentication configuration alongside params, headers, and body data. - Request execution now generates authentication headers and query values at send time, keeping manual headers and explicit auth settings separate. - Added tests for authentication request preparation, OpenAPI security inference, and auth persistence.

PrivyAPIStudio 0.1.4 focuses on a more modern macOS experience and smoother daily API workflows. Refined the main workbench, settings, and environment panels with a more modern macOS 26-style glass interface. Improved the AI Q&A panel with full Markdown rendering, clearer message attribution, and a better bottom input layout. Reworked environment and variable management with a current environment selector, add/rename dialogs, and separate variable tabs for Global, Project, and Environment scopes. Improved request parameter editing with stacked Path Params and Query Params plus a draggable divider. Cleaned up response viewing with a more focused Body/Headers layout and compact copy controls. Added a Help menu entry for configuring local LLM engines. Fixed multiple UI timing and layout issues, including nested variable dialogs and environment panel overflow.