01 · FIRST USE CASE
Start where the process has a stable core.
Good pilots combine recurring volume, measurable outcomes, explicit exceptions and a boundary the agent must not cross.
Agent Studio · Agent Runtime · governed execution
Enterprise AI agent platform for explicit tools, context and authority.
An enterprise AI agent platform should let teams configure agents, resolve their runtime capabilities, coordinate longer-lived work and control external actions from one operating model. Createting connects Agent Studio, Agent Runtime, Orchestrator and Mission Control under explicit tools, context, autonomy and approval boundaries.
✓
Voice and typed interaction
✓
Scoped models, tools, skills and knowledge
✓
Sensitive actions remain approval-gated
C
Simulated Orchestrator
Interactive agent runtime simulation
Ready
✦
ORCHESTRATOR
Agent Customer OperationsTools CRM · BrowserSkills Account ResearchKnowledge Billing PoliciesConditions Approval · Context
01TranscriptSpeech or typed input
02Route & contextModel, agent and resources
03Conditions & policyPermit, gate or deny
04Voice responseAuditable answer
Microphone permission requiredBrowser speech preview · no production execution · no customer data
WHERE CREATETING FITS
Use agents where work crosses systems, policies and teams.
Createting is built for recurring operational work that needs model reasoning, approved tools, business knowledge, conditions, human approvals and an audit trail—not for unbounded autonomy.
02 · GOVERNANCE
Control each resource—not only the agent.
Tools carry explicit restriction levels. Skills and knowledge bases remain scoped. Event, data, context, time, system, approval and reasoning conditions decide when a resource can activate.
NO_RESTRICTIONLOW_RESTRICTIONHIGH_RESTRICTIONEXPLICIT_DENY
03 · DEPLOYMENT FIT
Choose the runtime around the workload.
Use workspace-approved API endpoints, hybrid routing or private infrastructure according to data sensitivity, latency, availability and unit economics.
Model Router
policy routing across configured endpointsWorkspace Default
one explicit workspace-default route04 · BUYER VERIFICATION
Inspect what the platform claims to control.
- ✓ Model route is visible
- ✓ Resource scope is inspectable
- ✓ Approval ownership is explicit
- ✓ Execution evidence is retained
Recurring volumeTwo or more systemsClear risk boundaryMeasurable outcome
ONE MENTAL MODEL
Pathway = process. Agent = actor. Thinkway = method inside the agent.
Start with the work. Use a Pathway when the route can be defined. Use an agent when context must be interpreted. Give that agent a Thinkway when its working method should be reusable.
PathwayPROCESSRuns explicit steps, conditions, tools and approval gates.
Use a Thinkway when the same reasoning approach should be reusable inside the agent. A Pathway can hand an interpretive step to the agent and continue after a permitted result returns.
Not the reasoning actor.
AgentACTORInterprets context and does the work within configured autonomy.
The agent owns identity, model, instructions and configured capabilities. Agent Studio sets its autonomy and resource boundaries.
The actor can use a Thinkway; it is not the Thinkway.
ThinkwayMETHOD INSIDE AGENTStructures reusable reasoning and work stages inside the agent.
Nodes can organise instructions, tools, skills and conditions around stages such as gather, assess and verify. The agent applies that method to the case.
Not a separate agent and not an authorisation mechanism.
PathwayAgentusing a ThinkwayPolicy / approvalPathway continues
ARCHITECTURE
One control plane from intent to evidence.
Createting connects agent configuration, execution, coordination and human control so a workflow can move forward without losing its boundaries.
- 01
Agent Studio
Configure identity, instructions, tools, skills, knowledge, autonomy and runtime capabilities before deployment.
BUILD · AVAILABLE - 02
Agent Runtime
Execute configured agents through the enabled environments and surfaces for the selected workflow.
EXECUTE · QUALIFIED - 03
Orchestrator
Coordinate missions, agents, tools and dependencies while preserving scope and policy.
COORDINATE · CONTROLLED ROLLOUT - 04
Mission Control
Make execution state, approvals, evidence and artifacts visible to the people responsible for the outcome.
OPERATE · CONTROLLED ROLLOUT
REASONING MODEL
Choose the right kind of control for the work.
PATHWAYSDETERMINISTIC
Make the route explicit.
Pathways work like familiar workflow automation: connect predefined steps, conditions, tools and approval gates.
Use it for repeatable operations, approvals and predictable handoffs.
THINKWAYSAGENTIC
Let the agent reason within a boundary.
Thinkways structure reasoning and work stages inside an agent. The agent interprets context and acts; the Thinkway supplies a reusable method.
Use it when the next step depends on evidence and context.
Where do channels fit?
Voice, chat, browser, computer, API, webhook and schedule are execution surfaces. They are ways work reaches the runtime—not the platform's control model itself.
How does governance travel with an action?
Autonomy profiles, conditions and human approvals determine whether an action can proceed, needs review or remains outside the configured boundary. Evidence makes the result inspectable.
OPERATING PRIMITIVES
Context, capability and evidence stay connected.
The platform's value is not one agent or one channel. It is the contract around what an agent can know, do and prove.
Memory & context
Knowledge bases provide governed retrieval. Memory carries the configured context that an agent or mission is allowed to use.
- Scope
- Workspace, agent or mission context
- Boundary
- Access and write policy
Tools & extensibility
REST, MCP, webhooks, functions, custom tools and skills extend what agents can do without turning every action into an unbounded permission.
- Decision
- Tool, skill or knowledge action
- Control
- Autonomy, conditions and approval
Evidence & boundaries
Execution state, approvals, artifacts and audit evidence make outcomes inspectable while tenant and context boundaries remain explicit.
- Visible
- Status, decision and artifact trail
- Boundary
- Tenant and execution context
Availability, rollout status and enforcement depth vary by deployment. Architecture review makes those boundaries explicit before a pilot.
Compare my workflow →AGENTIC AI PLATFORM · PRODUCT TOUR
Build, coordinate and operate agents through one governed system.
Configure agents and their autonomy in Agent Studio. Use Pathways for rule-based workflows and Thinkways for reusable reasoning methods inside agents. The Orchestrator and Mission Control coordinate the wider work.
01 · ORCHESTRATOR
Turn a business goal into coordinated agent work.
Plan work, delegate bounded tasks, invoke approved tools and execution surfaces, and keep activity, questions and artifacts connected to the mission context.
↳Multi-turn conversation and reusable context
↳Guarded autonomy and model routing
↳Artifacts, browser sessions and human questions
REPRESENTATIVE RUN · EXECUTION STRUCTURE IS PLATFORM-ACCURATE
CG
Resolve a duplicate payment request within the approved refund boundary.
✦
Orchestrator
I verified the transaction, matched the customer identity and checked the financial policy. The €48 refund is permitted and a reviewable CRM update is ready.
Policy boundary
CompleteTransaction evidenceVerified
Policy boundaryPermitted
Artifact and audit trailReady
02 · MISSION CONTROL
See autonomous work as accountable operations.
Track mission state, task ownership, dependencies, approvals, execution evidence and outcomes in one operational view.
REPRESENTATIVE MISSION BOARD · STATES AND FIELDS ARE PLATFORM-ACCURATE↳Incoming, generated, running and approval states
↳Drag missions between lifecycle stages
↳Inspect costs, results and governance events
Mission ControlOperational command layer
Running
42 on schedulePlans ready7awaiting start
Approvals2needs attention
At risk1deadline warning
Incoming
2
Customer Operations
HIGH
Enrich priority accounts
Research new enterprise leads and prepare a reviewable CRM brief.
Execution plan · 3 waves
Open mission →
Revenue Operations
MEDIUM
Review disputed invoices
Collect evidence and classify disputes before human review.
Budget policy attached
Review plan →
Auto-Generated
1
Finance Operations
HIGH
Recover failed onboarding
Example recovery mission prepared for repeated workflow failures.
Wave 1 / 3Inspect →
In Progress1
Research AgentHIGH
Prepare renewal brief
Usage, support and stakeholder signals are being assembled.
€2.48 / €6.00Pause
Pending Approval1
Finance AgentCRITICAL
Approve €640 account credit
Evidence is complete. Financial action exceeds the autonomous boundary.
Approval reason · amount above €100
Awaiting approvalReview →
Completed1
Support AgentNORMAL
Audit escalation quality
Resolution pack, lineage and evaluation results are available.
€3.12 totalResults →
03 · PATHWAYS & THINKWAYS
Build workflows. Give agents reusable methods.
Pathways connect predefined steps and conditions. Thinkways organise reasoning inside an agent: each node describes a work stage with instructions, tools, skills and entry conditions. The agent applies the method to changing context.
↳Forced, suggested and reference patterns
↳Versioning, variables and evaluation suites
↳Connected nodes with live configuration states
PATHWAY · RULE-BASED PROCESSPredefined workflow executionSTART · TOOL · DATA_SOURCE · SKILL · CONDITION · MISSION · HUMAN · FINISH
entry_node_id · nodes[] · edges[] · draft/publisheddelegates interpretive work → agent applies Thinkway
THINKWAY · METHOD INSIDE AGENTReusable reasoning workflow inside the agentUnderstand · gather · assess · verify · propose
thinkway_id · variables · evaluations · versionsREPRESENTATIVE PATHWAY
STARTRequest
→DATA_SOURCELoad account
→CONDITIONPolicy check
→HUMANApproval gate
→TOOLExecute update
→FINISHResult
Thinkway variables and runtime scope
approval_policyconfigured thresholdnumber | rule
evidence_gaterequired before actioncondition
fallback_ownerassigned principal or teamprincipal
runtime_scopemission / sessionscope
Evaluation suite and release gates
Policy compliancegate configured
Unsupported action handlingdeny or escalate
Grounding requirementapproved sources required
Escalation qualityevaluation suite attached
Version and release states
CURRENTPublished or deployment-candidate definitionselected
DRAFTUnpublished graph, variable or enforcement changeseditable
BASELINELinked evaluation suite and comparison referenceattached
Graph valid5 nodes5 edgesReady to test
04 · AGENT STUDIO
Configure identity, resources, conditions and runtime before deployment.
Define identity, instructions, variables, model routing and autonomy. Attach tools, skills, knowledge and Thinkways—the reusable reasoning methods used inside the agent. Configure resource conditions and tool restrictions, then validate runtime and deployment readiness.
REPRESENTATIVE AGENT CONFIGURATION · STRUCTURE IS PLATFORM-ACCURATE↳Tools with four restriction levels and activation conditions
↳Skills with auto-invoke, manual use and conditions
↳Knowledge bases, runtime environments and deployment gates
CO
Customer Operations AgentDraft · Last saved 2 minutes ago
01
Identity
Define how the agent appears and what operating role it owns.
Agent name
Customer Operations Agent
BlueprintCustom operational agent ⌄
Agent identity
Resolves bounded customer operations cases using approved knowledge, tools and escalation policies.
02
System Prompt
Set behavior, scope and non-negotiable operating constraints.
System instructions
You are a customer operations agent. Verify identity and evidence before selecting actions. Never execute financial actions above the configured boundary without approval.
03
3 variablesVariables
Expose reusable configuration without rewriting instructions.
approval_limit€100number
support_regionEUstring
04
GuardedAutonomy
Choose how independently the agent can plan and act.
06
ConfiguredRuntime, models and advanced capabilities
Configure model routing and enabled runtime capabilities such as browser and computer execution, voice, memory, subagents and knowledge access according to the deployment.
CONFIGURED MODEL CATALOG
Model routingModel Router · Auto RouteWorkspace Default · OpenAI GPT-4.1 · Claude 4.8 Opus · Kimi K2.6
Browser policyRead-only sessionDomain allowlist active
SandboxIsolated workspaceEphemeral environment
Voice channelEnabledStreaming session · interruption
Parallel tool callsEnabled · limit 3Conditions and policy per call
Memory write policyobservations_onlyNo unrestricted memory writes
Subagent depthMax 2Research Council · Knowledge Search
computer_use_enabledtrueweb_search_enabledtruemodel_council_enabledfalsesupervisor_enabledtrue07
2 attachedThinkways
Attach reusable agentic reasoning workflows with variables, execution constraints and evaluation criteria.
Customer Recovery ReasoningForced · production candidate
v12Evidence Before ActionForced · verified
v708
6 / 8 readyDeployment readiness
Confirm configuration, tests and approval ownership.
deployment_stageguardedeligible_for_productionfalsemissing_gates205 · CAPABILITIES
Tools, skills and knowledge bases — governed resource by resource.
Tools carry NO_RESTRICTION, LOW_RESTRICTION, HIGH_RESTRICTION or EXPLICIT_DENY. Skills can auto-invoke. All three resource types support event, data, context, time, system, approval or reasoning conditions.
9 resources7 active conditions
tools[]tool_id · restriction level · conditionsskills[]skill_id · invoke mode · conditionsknowledgebase_ids[]knowledge_base_id · retrieval · conditionscontext_configsources · top_k · similarity threshold · memoryTOOLS3＋ Add
↗
CRM Account Operationscrm_connector · read/write
ON◎
Browser Sessioncomputer_use · isolated
ON⌁
Ticketing APIcustom_tool · scoped write
ONtargettool_id · skill_id · knowledge_base_idconditiontype · description · configevaluationbefore invocation, retrieval or actionoutcomeallow · deny · require approval · escalateCONDITION BUILDERHuman Approval
Require an authorized role and a minimum number of approvals before the resource activates.Natural-language descriptionRequire approval from a support lead when a financial action exceeds the configured limit.
Parametersrole: SUPPORT_LEAD · approvals: 1 · channel: orchestrator · expiry: 3600s
✓Condition valid · evaluated before resource activation
Scroll to move through the platform
01 / 04