Knowledge drift
Answers must come from approved, tenant-scoped sources—not an unbounded model memory.
Resolve / L0 Agents
Automate the first contact without automating the risk.
L0 Agents use published workflows, approved tenant knowledge, and explicitly granted tools. When a request crosses a confidence, policy, or authorization boundary, the full support state moves to a person.
- Built for
- Support leaders, automation owners, and technology teams
- Answers
- What can L0 resolve—and what happens when it should not?
01 Versioned workflow 02 Authorized nodes The first-contact gap
A bot can answer. A governed resolver must know its boundary.
High-volume support fails when automation optimizes for reply rate instead of safe resolution. OpenOS separates answer generation, authorized action, and evidence-backed escalation.
Tool risk
A workflow node can execute only a persisted tool granted to that exact release and node.
Cold handoffs
An escalation is incomplete unless the reason, transcript, collected fields, and customer context travel together.
Mechanism
From intent to safe resolution
Understand
Classify the request and collect the required context using the published workflow.
Ground
Retrieve from approved knowledge and expose the source context used by the workflow.
Act or stop
Use an authorized tool only within its grant; otherwise follow the explicit escalation path.
Preserve
Write the conversation, workflow state, reason, and evidence into the shared support memory.
Operational control
Release controls operators can inspect
The safety boundary is a product surface, not a sentence in a policy.
Versioned workflow
Draft, validate, publish, and bind a specific workflow revision to the channel runtime.
Scoped knowledge
Tenant knowledge, import state, and source access remain separate from public model knowledge.
Node-level grants
HTTP tools fail closed unless the release, node, tool, and allowed action are explicitly linked.
Human fallback
Escalation is a designed branch with a reason and payload—not a generic apology message.
Demonstration scenario
A return-status request reaches the automation boundary.
Illustrative workflow using demonstration dataCustomer asks about a delayed return and provides the order reference.
Workflow checks approved policy and attempts the authorized status lookup.
The status is inconsistent and the workflow cannot safely promise a refund date.
Workspace receives the transcript, order context, failed lookup evidence, and escalation reason.
Measures
Define success before the workflow changes.
Eligible containment
Resolved eligible requests divided by the configured eligible request population.
Escalation completeness
Handoffs that include the required transcript, fields, reason, and customer context.
Tool failure rate
Authorized tool attempts that fail, time out, or return an invalid response.
Direct answers
What buyers ask next.
Is this another chatbot?
No. Conversation is the customer interface. L0 Agents are governed workflows inside a wider operating loop that includes human exceptions, quality evidence, and leadership decisions.
Can we decide which intents L0 handles?
Yes. Eligibility, workflow logic, knowledge, tools, and escalation branches are configured and released for the tenant.
Can L0 update another system?
Only through a configured HTTPS tool or provider path with the required secret reference and execution grant.
Tailored walkthrough
Bring one support workflow. Leave with a clear first loop.
A tailored walkthrough for Support/CX, operations, and technology leaders—using your channels, responsibilities, and success measures.
- 01Your current loopChannels, systems, volume, and the operational gap.
- 02The product proofResolve, escalate, audit, and decide in one scenario.
- 03A scoped first deploymentOne workflow with explicit evidence and success measures.
Prefer email? sales@openost.com