ServiceStack v10.1!

mythz · August 13, 2026, 12:51pm

ServiceStack v10.1 is our largest feature release in years, which continues to enhance your typed Request DTOs which remain the single source of truth for clients, schemas, UIs, AI tools, authorization and production documents.

Typed APIs in 15 languages

Add ServiceStack Reference now supports Go, Rust, Ruby and Zig, bringing end-to-end typed API integrations to 15 languages:

C#, F#, VB .NET, TypeScript, JavaScript, Python, PHP, Ruby, Swift, Java, Kotlin, Dart, Go, Rust and Zig.

Each integration includes native DTOs and an idiomatic Service Client with typed responses, authentication, structured errors, validation, AutoQuery, uploads, batch requests and more - without maintaining separate SDK libraries.

AI Chat v4

v4 includes a complete, full-featured modular AI Assistant which can be added to any .NET 8+ ServiceStack App that’s available at /chat.

AI Chat supports multiple commercial, open and locally hosted providers-including OpenAI, Anthropic, Google, Groq, xAI, Mistral, Ollama and other OpenAI-compatible models-while using your App’s existing authentication and data boundaries.

Highlights include:

Enterprise RAG with Gemini File Search
Specialized Agent Profiles and workflows
Persistent Projects and reusable Skills
Voice input, image and audio generation
Per-user history, files and media
Usage, token and cost analytics
A built-in MCP Server
Opt-in filesystem and code-execution capabilities

Turn existing APIs into safe AI capabilities

New API Tools let AI assistants progressively discover, understand and invoke your existing ServiceStack APIs using just three tools: api_search, api_describe and api_call.

Calls execute as the authenticated user, preserving your existing roles, permissions, validation, filters and business rules. Read operations can run immediately, while writes and destructive actions can require an editable, schema-generated approval form.

Selected tools can also be exposed to external AI assistants through the built-in MCP endpoint at /chat/mcp.

API Schemas and generated API UIs

Every API now has a portable JSON Schema and an executable, schema-driven UI:

/schema provides a searchable catalog of authorized APIs
/schema/{RequestDto} provides a complete API workbench
/schema/{RequestDto}.json exposes the portable contract

Existing descriptions, validation attributes, allowable values, references and input metadata automatically produce richer forms-without maintaining a separate frontend schema.

Complete AutoQuery CRUD Apps

The new /auto experience turns AutoQuery models into authorized, full-featured data applications with searchable grids and schema-driven Create, Edit, Patch and Delete forms.

References become live lookup controls, validation becomes immediate UI feedback, and authorization determines which models and operations each user can see. The same schemas also power editable approval forms when AI Chat proposes data changes.

PDF Studio and production PDF rendering

The new PDF lifecycle combines AI-assisted document design with deterministic production rendering using Typst.

PDF Studio can create or recreate documents from natural-language instructions and screenshots, provide live previews, generate strongly typed C# contracts, validate realistic fixtures and publish immutable, versioned artifacts.

Production only needs PdfFeature and Typst - no LLM, ChatFeature or AI provider is involved when rendering invoices, reports, labels or other documents.

Also included

ServiceStack.Azure has been modernized to use Microsoft’s latest supported Azure SDKs, including Azure.Messaging.ServiceBus.

Everything works with .NET 8+ ServiceStack Apps. API Schemas and AutoQuery Schemas are already included with Metadata, with nothing extra to install.

Explore the complete v10.1 release notes, screenshots, examples and setup instructions:

bgiromini · August 18, 2026, 3:07pm

Awesome work and something we are actively researching, especially around the Enterprise RAG.

mythz · August 19, 2026, 3:22am

ServiceStack v10.1.4 Released:

Durable Agents, MCP Two-Phase Confirmations & Scale Enhancements

ServiceStack v10.1.4 brings major scalability and reliability enhancements to AI.Chat and our built-in Model Context Protocol (MCP) server.

Why the update?

After running complex, long-running agent tasks that accumulated massive 200MB+ JSON message payloads which effectively broke browser rendering and database tooling - the message-thread storage and execution was re-architected into a durable, memory-efficient, and slice-based checkpointed background tasks designed for handling long running agent workloads.

What’s New in v10.1.4

Durable Agents & Resumable Execution - Multi-step agent tasks now run in background slices checkpointed to the database, surviving browser disconnects, tab closes, and page refreshes without losing state.
Automatic Context Compaction - Seamlessly continues long chats past model context limits by automatically summarizing older history at ~80% capacity while preserving instructions and visible audit history.
Windowed Message Loading - Keeps the UI responsive and fast even on conversations with thousands of messages via on-demand windowed loading and optimized database queries.
SSE with Long-Polling Fallback - Real-time stream updates over Server-Sent Events with transparent fallback to long polling when proxies block SSE.
MCP Two-Phase Confirmation Tokens - Secure human-in-the-loop approvals for mutating API calls over MCP using cryptographically signed, short-lived tokens (new default ApprovalMode).
New [Mcp] Attribute - Inject imperative agent instructions and prerequisite rules into DTOs without polluting your public OpenAPI descriptions.
New fetch_url Core Tool - Zero-dependency web reader that converts web pages into clean, token-efficient Markdown, stripping scripts and boilerplate.
New grep_search Core Tool - Fast regex/text search with file globbing that automatically skips binary files, node_modules, .git, and build folders.
Provider Compatibility Enhancements - Normalizes context projections for strict model families (e.g. GLM) and fixes schema validation on parameterless MCP tool calls.

Checkout the ServiceStack v10.1.4 Release Notes for the full details.

mythz · August 24, 2026, 9:16am

For @bgiromini and anyone else making use of the Gemini RAG support, can you try the latest version of AI.Chat that’s in the new v10.1.5+ that’s in the new pre-release packages.

See the vnext Gemini RAG documentation for the docs and screenshots of the new features.

Gemini RAG updates in AI Chat

Gemini RAG has been expanded from File Search into a basic document management platform complete with ingestion, retrieval, synchronization, and embeddable Website Assistants.

New features

Three deep-linkable workspaces - Explore, Import, and Assistants preserve the selected store, category, import mode, crawl, Assistant, and conversation in the URL.
Expanded imports - upload individual files or ZIP archives, import server folders, or crawl websites into Markdown.
Previewable folder imports - compare new, changed, metadata-only, unchanged, removed, skipped, and failed documents before committing an import.
Recurring imports - save folder imports, re-run diffs, avoid re-embedding unchanged content, track removed upstream documents, and prevent unexpectedly large deletions.
Trusted import roots - restrict non-admin folder imports to configured directories, including aliased paths, with symlink escape protection.
Two-stage web crawling - crawl into an inspectable Markdown workspace before sending anything to Gemini.
Configurable crawl boundaries - include/exclude paths, depth and page limits, query-string policies, supporting hosts, robots directives, canonical URLs, and ordered crawl rules.
Crawl transforms - apply saved regex transformations across extracted Markdown, inspect the generated directory tree, then hand the result to Folder import.
Versioned import.json configuration - define global metadata, nested directory overrides, path rules, skip rules, and transforms for folder and ZIP imports.
Structured metadata - categories, document type, status, locale, product, versions, tags, and canonical Source URLs.
Source URL templates - construct citation URLs using validated variables including {fullPath}, {path}, {pathNoExt}, {dir}, {filename}, {name}, {ext}, {category}, and {title}.
Metadata-scoped Explorer - compact facet filters, visible filter chips, category navigation, coverage reporting, upload states, and deep links.
Bulk metadata editing - update the current page or every matching document, with set, overwrite, clear, add, remove, and replace operations.
Store synchronization - detect missing local or remote documents, metadata differences, unmatched fields, and duplicate Gemini documents.
Filter-preserving RAG queries - “Ask about this” carries the current category and metadata filters into Gemini File Search.
Inspectable citations - responses retain source cards, retrieved excerpts, and canonical links for each grounded answer.
Improved upload feedback - live progress, direct links to uploading documents, automatic retry, and startup recovery for interrupted uploads.
Embeddable Website Assistants - publish a File Store or filtered document scope as a self-contained Shadow DOM chat widget.
Assistant behavior templates - Documentation Guide, Technical Troubleshooter, Customer Support, Developer/API Assistant, Product Advisor, Onboarding Guide, and Policy & Procedures.
Per-Assistant configuration - private system prompt, response detail, grounding requirements, citations, suggested questions, fallback text, optional Gemini model, and automatic opening behavior.
Customizable widget appearance - Light, Dark, Nord, Matrix, and Soft Pink presets with independently editable colors, borders, typography, message bubbles, and launcher styles.
Hosting controls - exact and wildcard origin allowlists, request rate limits, stable deployment IDs, publish/unpublish, and presentation-only embed overrides.
Full widget experience - streaming Markdown, clickable citations, suggested questions, keyboard shortcut, fly-in animation, maximized view, persistent browser sessions, and clearable conversations.
Conversation review - retained conversations show referrer domains and pages, user-message counts, Markdown responses, citations, and navigation between customer questions.
Complete lifecycle management - publish, unpublish, regenerate deployment ID, archive, restore, and permanently delete Assistants.
Safe cascading deletion - deleting an Assistant or File Store shows the complete impact and requires typed name confirmation. File Store deletion removes associated documents, imports, Assistants, conversations, messages, and remote Gemini data.