CREATE
Agent Designer
Define an agent’s inputs, prompt or execution behavior, and output contract. Test it before publishing it for reuse.
- Prompt and external agents
- Widget, JSON, API, and web-session inputs
- Published and draft lifecycle
Gilentic Agentic Ecosystem
Build agents.
Connect workflows.
Run real work.
A visual platform for designing reusable AI agents, connecting them into governed workflows, and publishing useful experiences for people and systems.
Visual designHuman checkpointsInteroperability and scalingSharing and reuse
Gilentic Agentic Marketplace
Discover agents ready to become workflow building blocks.
Browse reusable capabilities from Gilentic and third-party providers, review their input and output contracts, and bring approved agents into your organization’s workflows.
One product, a complete path
From an agent idea to an operational workflow
Design each capability once, reuse it as a workflow node, and control how people and systems interact with the result.
ORCHESTRATE
Workflow Designer
Use agents as nodes, connect compatible inputs and outputs, and make human participation an intentional part of the flow.
- Visual node-based design
- Widget and JSON handoffs
- Interactive and automatic steps
REUSE
Libraries
Organize client-built, published, and shared agents and workflows so teams can find approved building blocks quickly.
- Agent and workflow libraries
- Client and shared catalogs
- Clear ownership and status
Designed for real interaction
Automation when it should run. A person when one is needed.
Gilentic workflows distinguish between machine-ready JSON and human-facing widgets. Automatic outputs can move directly to the next compatible agent; widget outputs can be displayed, logged, and held for user input.
See how the product works →01
CollectScreen, API, fixed value, JSON, or authenticated web session
02
RunA published agent performs one focused unit of work
03
RouteContinue automatically with JSON or pause with a widget
04
GovernPublish, assign access, and keep the workflow understandable
WebSession input
Bring authenticated webpages into an agent run
A WebSession input opens the requested URL, waits for the user to complete authentication, and makes that approved browser session available to the agent. It is an input type and not a special agent type, so it fits naturally beside other prompt inputs.
Reference deployment architecture
A control plane for governance. Independent workers for execution.
Gilentic separates the product experience and run coordination from the infrastructure that performs agent work. This keeps design, authentication, configuration, and run history consistent while allowing execution capacity to scale independently.
Experience layer
The Gilentic frontend provides the designers, agent testing, and run history used by builders and operators.
Control plane
The control plane handles authentication, agent configuration, authorization, and run coordination. It creates the identifiers that keep work traceable throughout the platform.
Execution plane
Independent workers execute agent runs using authorized configuration supplied by the control plane. Capacity can scale separately as demand changes.
Durable state
Agents, runs, events, results, captures, documents, and other outputs are retained so every execution can be reviewed and governed.
Centralized monitoring provides operational visibility and alerts. Optional interoperability interfaces can expose agent capabilities to other systems while preserving the same governed Gilentic run model.
Deployment model
Use the simplest execution path that fits the work.
A focused agent run and a multi-step workflow share the same identity, configuration, event, and result model. The execution path changes only when the workload needs coordination beyond one independent worker.
Queue-based isolationIndependent worker scalingTraceable run stateCentralized monitoring
FOCUSED AGENT RUN
Queue one execution
The control plane records the run, queues its identifier, and lets an isolated worker execute and report events and structured output.
Control plane→Queue→Worker
MULTI-STEP WORKFLOW
Coordinate durable steps
Durable orchestration coordinates retries, checkpoints, and parallel or converging paths while starting the appropriate workers and preserving workflow state.
Control plane→Orchestrator→Workers
Built for an ecosystem
One workspace for builders, operators, and providers
Create agents and workflows visually, then test the behavior before publishing.
Control which published capabilities are available to each client and user.
Offer reusable agents with defined contracts while retaining control of provider-hosted execution.
Start with one useful agent