Layi Docs

Create Agents in Four Ways

Import existing JSON, draft a brief, clone from the marketplace, or duplicate a proven worker

Last updated: December 2025

Pick the workflow that matches your org

Whether you are migrating legacy JSON exports, rapidly prototyping with an LLM brief, or standardizing on marketplace templates, the agent builder keeps every path in a single modal (see screenshots). Each flow outputs a complete agent with prompts, tools, voice settings, and governance metadata.

JSON / file import
Paste exported JSON or upload a .json file to recreate an agent exactly as it ran elsewhere.
Brief to agent
Write a natural-language brief; Layi’s LLM scaffolds the prompt, tools, and KPIs automatically.
Marketplace clone
Browse public agents, inspect their goals, then clone into your org with one click.
Duplicate existing
Copy an internal agent, tweak prompts or tooling, and publish faster than starting from scratch.

Import JSON or upload a file

Choose Import from JSON or Import from file at the bottom of the modal. Paste the exported payload or choose a saved .json file. We validate schema, preview the agent name, and show which tools/knowledge will be recreated before you clickNext.

Import agent from JSON modal
  • Safe replays

    Imports never overwrite existing agents; duplicates append a numeric suffix until you rename them.

  • Secrets

    Tool credentials are intentionally stripped—reattach secrets through the Tool Gateway after import.

Create from a brief

Toggle Uncheck to create basic off, then fill in the brief fields shown in the screenshot: role, goals, steps/workflow, and tool hints. When you hit Create, Layi’s backend brief builder uses your org’s approved templates to generate a production-ready agent with KPIs, evaluation hooks, and default memory attachments.

Create new agent modal
  • Agent goals = KPIs

    Each line you enter becomes a measurable goal in the agent dashboard, letting supervisors track completion rates.

  • Suggested steps

    Steps serve as the agent’s executable checklist; they surface in the run timeline as a guide and unlock troubleshooting telemetry.

  • Tool hints

    List the tools or data sources you expect; the builder auto-links eligible MCP integrations and flags missing access.

Clone from the marketplace

Use the Explore tab to browse public agents. When you click Use Agent, we duplicate its configuration into your org, strip org-specific secrets, and map optional tools to the closest equivalents you have installed.

Compliance friendly
Every marketplace agent surfaces its creator, last audit date, and required data categories so your review board can approve it quickly.
  • Preview prompts

    Inspect the system prompt, goals, and steps before importing so you know exactly what the agent will do.

  • Version locking

    Cloned agents freeze at the version you imported; subscribe to notifications if the author ships updates you want to merge later.

Duplicate an existing agent

From My Agents, open any worker and click Customize → Duplicate. This keeps prompts, tools, voice routing, and knowledge attachments intact so you can experiment without touching the production instance.

  • Safe sandboxing

    Duplicates land in draft state until you publish; run test threads and adjust KPIs privately.

  • Template publishing

    Once dialed in, publish as an internal template or back to the marketplace if you want to share with other orgs.

After you create the agent

All four creation paths drop you into the agent workspace (Model, System Prompt, Tools, Voice & Calls, Integrations, Knowledge Base, Playbooks). Review each tab before publishing or routing live calls.

Agent configuration workspace
  • Goals dashboard

    KPIs automatically populate the agent performance board so you can benchmark runs immediately.

  • Suggested steps telemetry

    The run timeline compares actual actions to your suggested steps and flags drift.

  • Voice routing

    Assign or recycle phone numbers, configure hold prompts, and test mute/unmute from the same screen.

  • Modular configuration

    Swap custom models, enable/disable tools, attach knowledge files, configure playbooks, and wire integrations without leaving the tabbed workspace.

Navigate the agents hub

The left navigation surfaces Explore, My Agents, and New Agent so operations teams can review production workers, spin up new briefs, or browse marketplace templates side-by-side with billing, integrations, and API key controls.

Agents hub overview
  • Org-wide visibility

    The hub lists every private agent plus public marketplace options so reviewers can cross-check coverage gaps.

  • Context-aware prompts

    Suggested tasks update based on the active org and selected agent, keeping launch instructions consistent.

Select any agent for quick actions

Clicking an agent tile opens a modal with Chat, Customize, and Publish as Template options. Use it to jump into testing, edit prompts, or share a vetted worker with other teams in a couple clicks.

Agent detail modal
  • Chat

    Opens a thread with the live agent using production routing so QA can validate behavior with real tools.

  • Customize

    Drops you back into the full workspace where you can update voice profiles, secrets, playbooks, or custom models.

  • Publish as template

    Share a hardened configuration to your marketplace catalog without exposing organization-specific secrets.

Run agents from the dashboard

The dashboard keeps start controls, attachments, suggested prompts, and playbooks in one place so supervisors can launch workflows and observe KPIs while switching between run modes (builder, capability test, or real task).

Agent dashboard
  • Ready-to-run

    One-click Start launches a fresh conversation using the agent’s latest configuration and routing plan.

  • Suggested prompts

    Pinned prompts document best-practice handoffs so analysts can trigger common workflows without rewriting context.

  • Playbooks

    Pre-built, variable-driven runs make it easy to schedule recurring jobs or multi-step escalations.

Tools Gateway: 500+ secure integrations

Layi operates a zero-trust Tools Gateway so orgs can safely connect more than 500 pre-integrated services. Each connection lives in aprofile that stores your credentials, rate limits, and audit metadata; once a profile is saved, the tool appears inside the agent’s Tools tab for assignment.

Connect and preview tools
  • Browse existing profiles

    Pick from saved org profiles so sensitive secrets never leave the vault and can be reused across agents.

  • Create new connections

    Spin up a new profile whenever you need a fresh credential set (per agent, per environment, or per region).

  • Tool visibility

    View the available actions before connecting so you know exactly what capabilities the agent will receive.

Create a profile and add secrets

Profiles separate identifiers (name, description) from the sensitive API keys. The gateway enforces secret masking, validation, and rotation policies before enabling the tool for assignment.

Create profile modal
Profile secrets entry
  • Name profiles clearly

    Use environment or workload hints (e.g., ‘Twilio - Ops Duty’) so admins know which credential powers each agent.

  • Vault-backed secrets

    API keys are encrypted, scoped to the profile, and never exposed to the agent UI or run logs.

  • Multiple credentials

    Create as many profiles per service as you need; assignment happens per agent so conflicts are avoided.

Assign tools to an agent

After a profile is saved, choose which actions the agent can call. Tool Gateway enforces least privilege—only the checked tools and profiles show up during runs.

Configure tools modal
  • Fine-grained selection

    Pick specific endpoints (trigger, action, analytics) so specialists get only what they need.

  • Version-aware

    Tool metadata shows version, parameter requirements, and audit dates so compliance teams can review access.

  • Instant availability

    Once saved, the tool slot lights up in the agent workspace; attach it to any step or playbook immediately.

Add custom APIs and legacy systems

If a service is missing from the catalog, upload its OpenAPI JSON (file or URL) or register a Custom MCP endpoint. Layi’s internal team reviews submissions, publishes the tool to your org’s catalog, and wires it through the same Tool Gateway guardrails.

Tools catalog
  • OpenAPI uploads

    Provide the spec JSON or URL; the gateway introspects endpoints and generates typed actions automatically.

  • Custom MCP

    Bring internal FastMCP services online with the Add Custom MCP button—perfect for proprietary data planes.

  • Legacy support

    For non-API systems, contact Layi with the service name; we’ll recommend relay options (RPA, email bridge, or hybrid connectors) and add it for you.

Reliability & Governance