Build and run agent solutions.
The managed operating system for autonomous work. Compose agents, capabilities, knowledge, and workflows. Noorle handles execution, memory, recovery, and policy.
- Orchestration
Agent loops and durable workflows
- Knowledge & memory
Context that carries across runs
- Tools & compute
Capabilities and execution environments
- Governance
Identity, policy, and a journal
Your prototype works. Now comes production.
Running an agent product means coordinating tools, keeping context, recovering from interruptions, and bringing people into the work.
Noorle provides that operating layer. Your team defines the solution, connects its resources, and sets the rules.
Explore a support workflow from configuration to a completed ticket.
Build your first agent“Investigate ticket TKT-482 and close it after review.”
Define the work. Bind the primitives.
- Agent
- Support specialist
- Knowledge
- Support playbook
- Capabilities
- Ticketing · Files · Memory
- Workflow
- Reviewed ticket closure
Your instructions and resources, assembled into one solution.
The runtime coordinates the work.
- Read ticket
- Ticketing connector
- Retrieve context
- Knowledge retrieval
- Draft a resolution
- Support agent
- Request closure
- Workflow awaiting approval
Context, tool calls, and the workflow stay attached to this run.
A human decision fits into the run.
- Proposed action
- Close TKT-482
- Policy
- Human approval required
- Workflow
- Paused with state retained
- Reviewer
- Can approve or deny closure
The runtime holds the work until the reviewer decides.
The work completes. Context carries forward.
- Ticket
- Closed after approval
- Resolution note
- Saved to workspace
- Useful context
- Stored in agent memory
- Run record
- Actions and usage recorded
The result is available to your team, with a record of the run.
Compose the solution. Choose how it runs.
Agents, Gateways, and Workflows share the operating foundation. Use them together to deliver your product through the interface your users need.
Agents
Compose an agent from capabilities, knowledge, and policy. Noorle runs its loop and holds its memory between runs.
Run an agentGateways
Expose a set of capabilities over MCP. External clients can call your tools through the same admission boundary.
Publish a gatewayWorkflows
Publish a Program for repeatable work. A run can wait for approval, survive a restart, and pick up where it paused.
Build a workflowThe primitives, already connected.
Tools, knowledge, memory, models, and automation work within the same runtime. Assemble the pieces your solution needs, with shared identity and state.
Explore the platformCapabilities & compute Built-in tools, plugins, and connectors
Fourteen built-ins include Files, Web Search, HTTP Client, Knowledge Retrieval, Code Runner, Sandbox, Computer, Browser, Plugin Builder, Skill Builder, Workflow Builder, Agent Builder, Memory, and Decision. Add WebAssembly plugins or connect REST, OpenAPI, and MCP services.
Explore capabilities & computeKnowledge & memory Organizational knowledge and durable context
Retrieve relevant passages from your documents. Keep conversation history and summaries, and let agents store, recall, and update durable facts across runs.
Explore knowledge & memorySkills Reusable instructions for your domain
Package how your team does a job into reviewable instructions that an agent loads when needed. Keep specialized ways of working available across conversations.
Explore skillsModels & execution Model selection, retries, and recovery
Choose a supported model or use Smart routing. Noorle handles model retries and fallbacks, records each attempt, and runs tools in the appropriate execution environment.
Explore models & executionWorkflows & automation Repeatable work with durable waits
Publish a Program as a versioned Workflow. Wait for a timer, approval, signal, or callback. Use schedules, watchers, and check-ins to start work.
Explore workflows & automationAutonomy runs within your rules.
Every side-effecting call crosses one admission boundary. Identity, capability grants, policy, and an append-only journal make the work attributable.
Inspect the trust model
tickets.read("TKT-482")- Identity
- agent/support-triage
- Capability grant
- read · granted
- Policy
- Allow ticket reads
- Journal
- Read allowed
Allow The connector can fetch the ticket.
tickets.close("TKT-482")- Identity
- agent/support-triage
- Capability grant
- close · granted
- Policy
- Require approval to close
- Journal
- Approval requested
Pause The ticket stays open while the run waits for a reviewer.
tickets.delete("TKT-482")- Identity
- agent/support-triage
- Capability grant
- delete · missing
- Policy
- Skipped after grant refusal
- Journal
- Missing grant recorded
Deny The runtime refuses the call. The ticket stays in place.
Pay for what your agents run.
Model use, calls, services, and retained data draw from one balance. Estimate a workload with the public rate card.
Example call charges
- 1 run start
- $0.000050
- 20 tool dispatches
- $0.001000
- Call subtotal
- $0.001050
Model, service, and storage charges apply separately.