In development. Not on npm yet.

Build AI agents
from files.

Write your agent as a folder of Markdown and TypeScript. Plico will run it, with the whole harness around it already built.

THE FILES ARE THE INTERFACE An agent, taking shape
01 / YOU WRITEA few files.
support-agentlocal project
Define its jobMARKDOWN
# Support agentHelp the team triage support requests.Follow the triage skill.Create a ticket when action is needed.
plico.config.tsevals/memory/artifacts/
Your logic lives hereMarkdown + TypeScript
Just files. Keep them in Git.
02 / PLICO HANDLESThe harness.
plicoaround your agent
Agent loop
Tool execution
Event log
Studio + API
Support agentfrom support-agent/
ready to take shape

Your files become its behaviour.

YOU

The team can’t connect to the VPN. Can you open a ticket?

create_ticket"Team VPN connection issue"
AGENT

Created ticket TCK-104 for the VPN issue, with the details your team needs.

Waiting for your files—
You define the behaviour.Plico runs the project.
ILLUSTRATIVE RUNIn development
01 / 04

Start with what your agent should do.

Plain Markdown gives your agent a job. Your instructions stay readable, reviewable and yours.

You write the agent. Not the plumbing.

A model on its own can only answer. To do real work it needs a harness: a loop, tools, approvals, logs, memory and tests. Most teams build all of that by hand. Plico will do that work for you.

Without Plico

  • Write the agent loop
  • Wire up tools and validation
  • Build approval and resume logic
  • Add logging, storage and a debug UI
  • Then write the agent

With Plico

  • Write the agent

An agent is just files.

Readable, reviewable and kept in Git like the rest of your code. These come from the example project in the repository.

agent.md
# Internal Ops Agent

Use the local Markdown instructions
as the source of truth.
skills/triage.md
# Triage

Review incoming work items, identify
the smallest safe next step, and
record the outcome.
tools/create-ticket.tool.ts
export default {
  name: "create_ticket",
  description: "Create a support ticket.",
  capabilities: ["ticket:write"],
  approval: { required: false },
};

One folder, from laptop to production.

The plan is that nothing about the folder changes as the agent grows.

  1. TerminalRun and test the agent while you write it.
  2. Local StudioWatch each step, then replay a run to debug it.
  3. Your appCall it over HTTP or drop in a React chat UI.
  4. ProductionScale to many workers, with the same folder.

Where it stands

Plico is early. Here is what is real and what is not yet.

Works today

  • Project folder and validation
  • Scripted runs and evals
  • Readiness checks
  • Event log, HTTP API, local Studio

Up next

  • Real model runs with streaming
  • Approvals that survive a restart
  • Artifact versions and diffs
  • Skills and TypeScript tool authoring

Planned

  • Published to npm
  • MCP tools and memory
  • AG-UI and a React kit
  • Postgres and multi-worker production