Developer preview · 0.1 on npm

Bring prompts and tools.
Plico handles the harness.

Write the job in Markdown and the tools in TypeScript. Plico gives your agent durable runs, an event log, approvals, artifacts, evals and a local Studio, inside your backend or on its own.

Get startedRead the docs
get startednpm create plico@latest my-agentFollow the quickstart

agent.md

# Support agent

Help the team triage
support requests.

skills/triage/SKILL.md

# Triage a request

1. Read the issue.
2. Decide what's needed.
3. Open a ticket.

tools/create-ticket.tool.ts

defineTool({
 name: "create_ticket",
 handler: async () =>
  ({ ticketId }),
});
plico (Latin): I fold. Your prompts and tools, folded into an agent.

See it run in a minute.

An example ops agent in the local Studio: chat with it, approve a tool call, replay the run, review the file it wrote, run evals and check readiness.

plico-launch.mp40:59

01project.loaded

An agent is just files. The harness comes with Plico.

Describe the job in Markdown, teach a workflow with a skill, give it an action with a TypeScript tool. Plico loads the folder and handles the run around it.

my-agent/Define its job
# Support agent

Help the team triage support requests.
Follow the triage skill.
Create a ticket when action is needed.
plico eventsexcerpt
  1. 01run.startedsupport-agent
  2. 02instructions.composed2 sources
  3. 03tools.discovered1 tool
  4. 04model.request1 tool offered
  5. 05tool.callcreate_ticket
  6. 06artifact.file.writtentickets/new.md
  7. 07tool.result{ ticketId: "TCK-1042" }
  8. 08assistant.outputCreated TCK-1042.
  9. 09run.completedcompleted

02run.started

One run, one event log, many views.

Every run writes the same durable log. The CLI, the API and Studio all read it, so what you see in one place is exactly what happened.

CLI

$ plico events

Print any run's events from the terminal, or replay a whole run offline.

HTTP API

$ plico serve

A thin native API over the runtime and the event store, for your own tools.

Studio

$ plico serve --studio

Watch runs, approve tool calls, chat with the agent and browse artifacts.

03tools.discovered

The harness, included.

You write the files. Plico supplies what usually takes a framework, a database and a few weekends.

Tools

Typed TypeScript handlers with schemas, capabilities and idempotency keys.

Approvals

Pause a tool call until someone approves it in Studio or over the API.

Artifacts

A versioned workspace with diffs and exports for what a run produces.

Threads

Multi-turn conversations stored next to the runs that belong to them.

Evals

Scripted scenarios in evals/ that check the agent without an API key.

Readiness

A checklist that tells you what is missing before you ship the agent.

04readiness.checked

Where Plico is today.

Plico is a developer preview, published on npm at 0.1. Here is what exists and what does not.

Works today

  • Project folder and validation
  • Scripted runs, evals and readiness checks
  • Event log, HTTP API and Studio
  • Provider-backed chat with streaming, approvals and cancel
  • Anthropic, OpenAI, Google or any OpenAI-compatible model
  • Artifact workspace with versions, diffs and exports
  • Threads, and embedding in your own backend
  • OpenTelemetry tracing

Up next

  • MCP and external tools
  • Context compaction
  • Eval scorers

Planned

  • Agent memory
  • Multi-agent workflows
  • AG-UI and a React kit
  • Postgres and multi-worker production

05run.completed

Your agent is one folder away.

Scaffold a project with one command and run it with a scripted model. No API key needed.

Get startedView on GitHub

Agents: read /llms.txt, or send Accept: text/markdown to any docs page.