Tealfabric
AI

Vendor intake assistant

Policy-bound · Skill: vendor-intake-triage

Ask about vendor intake, approvals, or missing evidence…Send

Simulated conversation — integrations execute only after human approval

“Finally an agent that runs inside our ops workflows—not another chat window on the side.”

BizOps lead, mid-market B2B

Platform

AI, Agents, and MCP — inside your governed runtime

Policy-bound Trace AI with tenant skills, human approval, and integration execute—on the same boundary as Process Automation and Datapools.

Book a demoSee HITL workflow

Trace AI is not a disconnected copilot. It is policy-bound operational intelligence inside the tenant your BizOps team already runs.

Most teams feel pressure to add AI on top of email, spreadsheets, and integration sprawl—but copilots that cannot see processes, Datapools, or governed connectors create more risk than value.

Tealfabric agents load tenant skills, work on granted operational data, propose actions, and pause for human approval before anything executes under policy.

  • Triggered from chat, WebApp, process, scheduler, or event—the same agent everywhere
  • Automatic model routing so ops teams are not tied to a single LLM release cycle
  • Query and update granted Datapool and integration data without sidecar copies
  • Tenant SKILLS/ playbooks your ops team publishes and governs
  • Human-in-the-loop before risky execute_tenant_integration calls
  • Configurable MCP guardrails, sandbox profiles, and audit trail on every action

SKILLS implementation guide for ops playbooks

Entry points

One agent runtime. Five ways to start the same governed workflow.

Coherence across channels—not a separate copilot per surface. The skill, guardrails, and audit trail stay identical whether the session began in chat or from a process event.

  1. You: What is blocking vendor intake?
    Agent: 3 records in pending_approval…

    Human chat

    Ops leads ask questions, triage exceptions, and approve proposals in conversation.

  2. POST /webapp/vendor-intake
    → skill: vendor-intake-triage

    WebApp submit

    Public intake surfaces trigger the same skill-backed agent path as internal chat.

  3. step: agent_assist
    → propose → HITL → continue

    Process step

    Process Automation invokes agents mid-flow—with agent_wait and shared audit.

  4. cron: 0 8 * * 1-5
    → weekly intake review

    Scheduler

    Recurring checks—SLA reviews, stale record sweeps, renewal prep—run on a cadence.

  5. event: sync_failed
    → retry under policy

    Integration event

    Monitor alerts and connector events wake agents to summarize, retry, or escalate.

Live conversation

From Monitor alert to governed retry—without ad hoc scripts.

An integration engineer responds to a failed SAP sync. The agent triages Monitor history, checks Datapool state, proposes a policy-bound retry, waits for approval, executes under guardrails, resumes the paused process, and links the full trace to the incident record.

AI

Integration recovery assistant

Policy-bound · Skill: failed-sync-retry

Ask about sync failures, Monitor alerts, or governed retries…Send

Simulated conversation — integrations execute only after human approval

Capabilities

Operational AI with guardrails your security team can actually configure.

Every capability runs inside the governed operations runtime—Datapools, Process Automation, integrations, and audit in one boundary.

  1. chat · webapp · process
    scheduler · event

    One agent, many entry points

    Triggered by human, WebApp, process, scheduler, or event.

    The same policy-bound agent can start from a chat message, a WebApp form submit, a Process Automation step, a scheduled run, or an integration event—coherence across channels instead of separate copilots per surface.

  2. task → best model
    no playbook rewrites

    Right model, automatically

    Each agent uses the most suitable language model for the task.

    Tealfabric routes agent work to the appropriate model behind the scenes. Ops teams do not chase the release cycle—capability upgrades apply inside the tenant without rewriting playbooks.

  3. SELECT datapool.*
    policy: granted

    Query and update granted data

    Read and write operational records the tenant policy allows.

    Agents work against Datapools, process context, and integration results in place—no copying supplier lists into a sidecar database or replicating CRM rows into a chat context window.

  4. SKILLS/vendor-intake
    ops-owned playbook

    Skills your ops team owns

    SKILLS/ playbooks in the tenant for repeatable agent workflows.

    BizOps and tenant admins publish operational playbooks—exception handling, intake triage, renewal checks—that agents load by policy. Your team builds the workflow logic; the runtime enforces guardrails.

  5. request_human_input
    awaiting approval

    Human-in-the-loop by design

    Approvals, clarifications, and error routing when judgment matters.

    request_human_input dispatches to the right person when policy requires it—approve an integration execute, confirm a bulk update, or handle an ambiguous exception. Agents assist; humans decide on risky paths.

  6. MCP guardrails: on
    execute: policy-bound

    Safeguards you configure

    Coherent guardrails and tenant overrides on tools and execute scope.

    MCP guardrails, integration execute flags, and sandbox profiles combine into one policy boundary. Security and ops leads define what agents may propose versus run—secure by design, not bolted on after launch.

  7. agent_wait → execute
    same tenant loop

    Same tool loop as processes

    agent_wait, execute_tenant_integration, and process steps in one tenant.

    Agents share the execution model with Process Automation—wait for events, call integrations under policy, and hand results back to the flow. No second automation runtime to govern.

  8. propose · approve · run
    full trace logged

    Full action trace

    Every propose, approve, and execute step is logged.

    Audit history captures what the agent read, which skill it loaded, what it proposed, who approved, and what ran. Post-incident review and compliance questions get answers from one trail—not chat exports.

  9. Monitor: sync_failed
    agent triage

    Monitor-aware responses

    React to integration failures and operational signals.

    Agents can be invoked from Monitor alerts or process exceptions—summarize the failure, suggest retry or escalation, and route to HITL before any execute_tenant_integration call.

Who it is for

Tenant admins, BizOps leads, procurement ops, and security stakeholders who need AI on operational data and integrations—not generic copy generation in a sidecar chat.

Related workflow: Ops exception handling with human approval. Platform depth: Governance, Datapools, Integrations, Process Automation.

Ready to see it on your stack?

Put agents inside the workflows you already run—not beside them.

We walk through tenant skills, HITL approval, integration execute under policy, and the audit trail your security and ops teams need.

Book a demoStart a pilotGovernance depth