OrchStack Studio

Build Agents Visually \u2014 No Framework Required

A visual IDE for AI agents. Drag-and-drop canvas, prompt studio, knowledge manager, and evaluation suite \u2014 everything you need to design, test, and deploy agents without writing infrastructure code.

Visual Canvas

Drag, Drop, Deploy

Design complex multi-agent workflows by connecting nodes on an infinite canvas. No YAML. No config files. Just visual flow.

LLM Node
Connect any model — GPT, Claude, Gemini, or your own fine-tune
Tool Node
Wire up APIs, databases, and external services as callable tools
Condition Node
Branch logic based on output, confidence, or custom expressions
Human Gate
Pause execution for human approval, editing, or review
Sub-Agent
Nest entire agent workflows as reusable, composable blocks
booking-agent.canvas
User Input
WhatsApp message
LLM Router
claude-sonnet-4
Knowledge
services-kb
Condition
intent == book
Human Gate
approve booking
Prompt Studio

Craft, Version, and Test Prompts

A dedicated IDE for prompt engineering. Write once, test across models, version everything.

Version Control

Every prompt edit creates an immutable version. Roll back, compare diffs, and branch prompts like code.

A/B Testing Built-In

Run two prompt variants side-by-side, measure quality scores, and promote the winner with one click.

Template Library

Start from battle-tested templates for classification, extraction, summarization, conversation, and more.

Variable Injection

Dynamic variables like {{user.name}}, {{context.history}}, and {{knowledge.results}} auto-resolve at runtime.

system-prompt.v3
v3
saved
# System Prompt
You are {{agent.name}}, a booking assistant for
{{tenant.business}}.

# Knowledge Context
{{knowledge.services}}

# Rules
- Always confirm before booking
- Never share other customers' data
- Escalate to human if confidence < 0.7
Knowledge Manager

Upload, Index, and Retrieve Knowledge

Give your agents domain expertise. Upload documents, build knowledge bases, and get cited answers.

Document Upload

PDF, DOCX, TXT, URL

Drag-and-drop PDFs, DOCX, TXT files, or paste URLs. We handle the rest.

Automatic Chunking & Embedding

Configurable strategies

Smart chunking with overlap, automatic embedding generation, and incremental re-indexing.

Hybrid Search

Semantic + Keyword

Semantic vector search combined with BM25 keyword matching and cross-encoder re-ranking.

Source Attribution

Document-level citations

Every response cites its sources. Users see exactly which document, page, and paragraph informed the answer.

Evaluation Suite

Test Agents Before You Deploy

Confidence in production starts with rigorous testing. Run evaluations, track quality, and prevent regressions.

Test Case Libraries

Build reusable test suites with expected inputs, outputs, and edge cases. Import from production logs.

Automated Regression Testing

Run your entire test suite on every prompt change. Catch regressions before they reach production.

Quality Scoring

Automated metrics for relevance, faithfulness, coherence, and toxicity. Custom rubrics supported.

Side-by-Side Comparison

Compare two models, two prompts, or two versions on the same test set. Pick the winner with data.

eval-run-0042 · 24 test cases · 2 variants
Prompt v3 (Claude)Relevance 94% · Faithfulness 97%
Prompt v2 (GPT-4o)Relevance 88% · Faithfulness 91%
Code Mode

For Developers Who Prefer Code

Everything you build visually generates clean TypeScript code underneath. Switch between canvas and code at any time.

Full TypeScript type safety with the OrchStack SDK
Visual edits sync to code, code edits sync to canvas
Git integration — branch, commit, PR, and merge workflows
Export any agent as a standalone TypeScript project
View SDK Documentation
support-agent.tsauto-generated
import { Agent, Tool, Gate } from '@orchstack/sdk';

export const supportAgent = new Agent({
  name:    'Support Agent',
  model:   'claude-sonnet-4',
  prompt:  'system-prompt.v3',
  knowledge: ['services-kb'],
  tools: [
    Tool.calendar(),
    Tool.payments(),
  ],
  gates: [
    Gate.approval('booking.confirm'),
  ],
});

await supportAgent.deploy();
// → Live on all configured channels
Capabilities

Everything in One Studio

The complete toolkit for building, testing, and shipping AI agents.

Infinite Canvas

Pan, zoom, and build on an unlimited workspace. Organize complex multi-agent architectures with visual clarity.

Prompt Versioning

Every prompt change is tracked. Compare versions, roll back, and A/B test variants with built-in history.

Knowledge Bases

Upload documents, auto-chunk and embed, then query with hybrid semantic and keyword search with citations.

Eval Suite

Build test case libraries, run automated regression tests, and score quality across relevance, faithfulness, and more.

Code Export

Every visual workflow generates clean TypeScript. Export as standalone projects or deploy directly from Studio.

Git Sync

Full Git integration. Branch, commit, and merge agent definitions. Collaborate with pull requests and code review.

Frequently Asked Questions

Start Building Agents Today

From idea to deployed agent in minutes. No infrastructure to manage.

Now in early access · No credit card required