TicketIQ
Triage, duplicate folding and grounded answers for the service desk. Rules decide first, and a model is asked only when they are unsure.
- ServiceNow
- Jira
- Zendesk
- Freshservice
- Microsoft Teams
Agent Liya?
What are you waiting for
While you wait, someone else’s agent takes the work
Session
service-desk · ServiceNow-style ITSMFix the highest-priority ticket
Working in the ServiceNow-style ITSM
- service-desk / query Incident4 Incident records
- service-desk / get record ConfigurationItem CI-2001vpn-gw-01
- service-desk / read policy “VPN down for all remote staff: tunnel drops at sign-in”3 articles
- service-desk / propose change Incident INC0010041held for your approval
Waiting for your approval
Incident INC0010041
- AssignmentGroup
Service Desk→ Network Operations- State
New→ In Progress
- Web People sign in to the Liyagent app in any browser and talk to the agents they may use. The Liyagent app
- Teams Build the Liya app package in the console and upload it for your tenant in the Teams admin center. Then people open Teams and start talking. Set up Teams
- Google Chat Create a Chat app in your Google Cloud project and point it at Liyagent. Then people message Liya from your Workspace domains. Set up Google Chat
- Terminal Install the liya CLI with one command and run liya try on the sample systems, or sign in to your own instance. Download liya
- Mobile People open the Liyagent app on their phone and add it to the home screen. Approvals reach them there too. The app on a phone
01 · Pre-built agents
Start with five. Build any other.
Each product is a set of agents for one part of the business and the systems behind it. They share one control plane with the agents you build, so every agent is governed, budgeted and audited the same way.
HR agents
Answers on leave and policy from the HR record itself, and onboarding tasks followed through for each new starter.
- Workday
- BambooHR
- SAP SuccessFactors
- Greenhouse
Agents for SAP
Maintenance, purchasing, sales, stock and supplier invoices in S/4HANA, people and time off in SuccessFactors, and Field Service dispatch. Within SAP’s API Policy.
- SAP S/4HANA
- SAP SuccessFactors
- SAP Field Service Management
- SAP Integration Suite
Liya for Salesforce
Answers from Salesforce records with the asking person’s own access, runs your Flows and Prompt Builder templates, and searches Knowledge. Every write waits for a person’s approval by default.
- Salesforce
- Salesforce hosted MCP
- Data 360
- Tableau Next
Liya Finance
Matches supplier invoices to their purchase orders, flags price and quantity variances beyond your tolerance, holds a mismatched invoice for AP approval, and answers spend questions.
- NetSuite
- Dynamics 365
- Ramp
- DocuSign
Liyagent platform
Build agents for any other system, or register the ones you have. Connect them through the built-in connectors, vendors’ own MCP servers, or anything over MCP, OpenAPI, OData, SQL or agent-to-agent.
Build your own50 built-in connectors
ServiceNowJiraZendeskFreshservicePylonincident.ioGrafanaWorkdayBambooHRSAP SuccessFactorsGreenhouseSAP S/4HANASAP Field Service ManagementSAP products (published APIs)NetSuiteDynamics 365Dynamics 365 Finance & OperationsDynamics 365 Business CentralRampDocuSignSalesforceMicrosoft TeamsSlackDiscordSharePointGmailGoogle DriveGoogle CalendarMicrosoft Entra IDOktaSQL databasesMongoDBMongoDB AtlasNeo4jNeo4j AuraRedisRedis CloudElasticsearchQdrantMetabaseAmazon RedshiftAmazon S3Amazon SNSAmazon SQSGoogle Cloud Pub/SubNATSAmazon BedrockOpenAICohereText chunker
Plus presets for vendors’ own MCP servers, among them GitHub, Confluence, Notion, PagerDuty, Salesforce, Dynamics 365 and SAP Integration Suite. Anything else connects over MCP, OpenAPI, OData, SQL or agent-to-agent. All connectors
02 · Control
Autonomy you set, tool by tool
Every system an agent reaches has a setting for its writes, and each tool can have its own. A new connection starts at approval required. Open it up as trust grows, and close it the moment you need to.
Example: a tool that closes ServiceNow incidents
Approval required
The call is held with its arguments exactly as they would be sent. A person approves, edits or rejects it, and the decision is audited. New connections start here.
Rules on the arguments only tighten it
- Close 3 incidentsruns
- Close 300 at oncewaits for a person
- Name a production hostdenied
Kill switches
Stop one agent, one channel or every agent at once.
Policies as code
Cedar policies decide which person or agent may do what.
Budgets
Spend is reserved before each model call and capped per agent, per team and per tenant.
Guardrails
PII redaction, prompt-injection blocking and denied topics, on inputs, outputs and tool calls, wherever you switch them on.
Red-team release gate
Built-in attack runs score each agent. Turn the gate on, and a new version ships only after a fresh run passes.
Hash-chained audit
Every step, approval and cost on a chain you can verify. Change or delete a row and the chain breaks.
Any model, with fallbacks
Anthropic, OpenAI, Gemini, Vertex AI, Bedrock, or any OpenAI-compatible endpoint, with a chain of fallbacks for each agent.
Grounded answers
Answers from your knowledge base carry a confidence band, so a weak match is shown as one.
03 · Your environment
Hosted by us, or run in your own
The same product either way. Postgres is the system of record, with a Qdrant vector index beside it, and your records stay in your environment unless you choose otherwise.
HostedRun by LiyagentHosting terms are agreed per customer during scoping.
Self-hostedIn your accountUnder an enterprise licence, installed with your team: Docker Compose on one VM, or Kubernetes, in your cloud or data centre.
ResidencyModel calls stay in regionA tenant’s model call is refused unless it can be shown to stay in the region the tenant names.
IdentitySSO and SCIMOIDC sign-in with Entra ID, Okta, Google Workspace, Keycloak or Auth0, and SCIM 2.0 provisioning from your directory.
ChannelsWhere people already askMicrosoft Teams, Slack, Google Chat, email, web chat and the workspace app, plus webhooks, schedules and agent-to-agent.
PreviewLiya MeetingAnswers questions asked in a meeting once consent is recorded. Off until you switch it on, and no audio is stored.
04 · Agents for SAP
SAP, by SAP’s rules
SAP’s API Policy limits how AI agents may call SAP’s APIs. Liyagent works within it in two ways, so you can start with the SAP you already run.
Directly to SAP’s APIs
An agent never calls SAP on its own. It proposes each call in plain words, the person who asked confirms it in Teams or web chat, and fixed code makes that one call. Creating documents stays off until your SAP Digital Access licence is acknowledged.
Through your SAP MCP gateway
Agents read through the MCP servers you deploy in SAP Integration Suite, the route SAP provides for AI agents. Anything that can change SAP is still confirmed.
Planner Is pump P-101 OK? Raise a notification if it needs one.
Liyain Microsoft Teams
Run this in SAP?
Create an SAP maintenance notification for equipment 10004711: “Inspect P-101 bearing. Vibration 7.1 mm/s, limit 9.0”
- Priority
- 2 · High
- Ticket
- P-101
This creates an SAP service and maintenance document, which may count toward your organisation’s SAP Digital Access licence.
Nothing is sent to SAP until you confirm. This expires in 15 minutes.
ConfirmCancel
05 · For builders
Build an agent for any system
The five products are starting points. In the console, an admin, or anyone your roles allow, builds an agent in one form, then connects it where people already work.
- InstructionsWhat it does, in plain words
- ModelAny provider, with fallbacks and a context window up to 2,000,000 tokens
- ToolsFrom the catalogue, or any MCP server
- Sub-agentsEach with its own instructions, context budget and limits
- LimitsBudgets, write settings and the guardrails you attach
- ChannelsTeams, Slack, Google Chat, email, web chat or the workspace
For developers
> liya apply -f ./agents --dry-run
Agents, policies, guardrails and triggers as YAML, applied with the liya command line. Also an MCP server, OpenAI- and Anthropic-compatible gateways, and a REST API.
06 · How we engage
From a first call to production
- 01
Demo
A working session on your use case, with the systems you run.
- 02
Pilot · 4 to 6 weeks
In your environment, against success criteria we agree up front.
- 03
Production
Roll out team by team, with the policies and budgets from the pilot.
07 · Questions
Frequently asked questions
What is Liyagent?
Liyagent builds governed AI agents that run in a company’s own environment and work with the systems it already has, through each system’s own APIs. Agents for IT service, HR, SAP, Salesforce and finance come ready to use, and you can build agents for any other system. Every agent shares one control plane for policies, budgets, approvals and audit.
Can I self-host Liyagent?
Yes, under an enterprise licence. We install it with your team in your own cloud account or data centre, with Docker Compose on one VM or on Kubernetes, and PostgreSQL as its system of record. Whether we host it for you instead is agreed during scoping.
Which AI models does it work with?
Anthropic, OpenAI, Google Gemini, Vertex AI and Amazon Bedrock, or any OpenAI-compatible endpoint such as Ollama, vLLM, OpenRouter, Groq or Azure OpenAI, including models on your own GPUs. Each agent has its own model chain, fallbacks and spend budget.
How does an agent decide what it may change?
Each connected system is read-only or not, and each write tool is allowed, held for a person’s approval or denied. A new connection starts at approval required. Rules on a call’s arguments can only tighten it, so closing three tickets runs while closing three hundred waits for a person.
How does Liyagent work with SAP?
Two ways, both built to work within SAP’s API Policy. Directly over SAP’s APIs, an agent never calls SAP on its own: the person who asked confirms each call in Teams or web chat. With SAP Integration Suite, agents can work through your own SAP MCP gateway. You remain responsible for your SAP licences.
How do we get started?
Try the sample systems on this site, then book a demo on your use case. A 4 to 6 week pilot runs in your environment against success criteria agreed up front, and production rolls out team by team.