crusbro agent studio

GUI · ZERO-CODE · TOOLS · SKILLS · LLM · DEPLOY

crusbro agent studio is a zero-code graphical builder for delivery teams: configure Tools, Skills, LLM endpoints and system prompts in the GUI, with Harness gates and freeze policy, and generate a complete agent program that deploys and runs on its own. The package already includes crusbro agent-OS and business integration code—no separate OS install before loading config.

Positioning

Zero-code is the default path: no hand-rolled scaffolding or JSON stitching—define tool contracts, skill graphs, model routing and gate tables in the GUI. One-click output is a self-contained agent program with crusbro agent-OS and business code inside, tagged with config_version, ready to run after environment credentials are injected.

LayerProduct / artifactRole
Buildcrusbro agent studioZero-code GUI config, validation, packaging, pre-release gates
PackageStudio outputagent-OS + business code + Harness config; deploy and run standalone
ScenariosIndustry agents / custom appsBusiness tools, acceptance criteria, integrations

GUI modules

ModuleConfigureBundle output
ToolsSchemas, side effects, idempotency, timeout, freeze rulestool_registry.json
SkillsMulti-step graphs, branches, tool orderskills/*.yaml
LLMEndpoints, routing, limits, fallbackllm_profiles.yaml
System promptRole boundary, deny list, revisionsystem_prompt rev
HarnessGates, freeze list, HITL rulesgates + freeze_policy

Build → deploy flow

  1. Import or new: blank, industry template, or clone from OPC / drawing / vision agents.
  2. Edit in GUI: connect Tools and Skills; forms for schemas and risk tags; prompt blocks with revision.
  3. Preview: sandbox tools + LLM; single-Run trace; deliberate failure paths for gates.
  4. Release gates: schema check, cycle detection, outbound tools must attach freeze policy.
  5. Package & deploy: manifest + config_version as a complete program; inject credentials per environment and deploy and run directly—no separate agent-OS install step.

Deployable package

Output is a complete agent program, not loose config fragments: entrypoint, crusbro agent-OS runtime, business integration layer, tool/skill registrations, Harness policy and version metadata in one package, runnable on its own. Credentials and master-data endpoints are injected per environment.

Consulting crusbro agent-OS Handbook