filepadai
sign inget started

Quickstart

Create a workspace, add operating notes, connect an agent, and review your first diff.

This quickstart is the shortest path to "Filepad makes sense to me" — one workspace, one agent, one completed run.

1) Create a workspace

Sign in and create a new workspace. Name it something you'd keep for months (client-alpha, content-pipeline, research-q3).

When the workspace initializes, 8 standard folders are created automatically:

.filepad/   agents/   skills/   memory/
sources/    uploads/  artifacts/  automations/

These are real directories. You can browse, open, and edit every file in them.

2) Add operating notes

Open agents/ and create an operating note such as research-agent.md. This is optional file context for the external agent you plan to connect. It is not a deployment target:

# Research Agent Notes

name: my-agent
model: gpt-4o

## Instructions

You are a research assistant. Work methodically and cite sources.
When in doubt, ask rather than guess.

## Scope

read:
  - agents/
  - skills/
  - memory/
  - sources/
write:
  - artifacts/
propose:
  - memory/

The Agent Access key controls what your external agent can touch. The Scope section is a human-readable note you keep aligned with that key. propose means writes are staged as diffs for your review.

3) Add content

Drop source documents into sources/. Add any reusable prompt modules to skills/. Write persistent context (known facts, preferences, project state) into memory/facts.md.

The agent will read all of this when it connects to your workspace.

4) Create an Agent Access key

Open the Access Keys tab in your workspace. Create a new key and grant it the scopes that match your operating note:

  • env:read — read the file tree, skills, memory, sources
  • artifacts:write — create outputs under artifacts/
  • files:propose — submit proposals to memory/ and other files
  • events.write — report durable activity back to Filepad

Copy the key secret — it's shown once.

5) Connect your agent

In your agent or pipeline, authenticate using the Agent Access key. Use the Agent Access API v1 to:

  1. GET /agent-api/v1/workspaces/:id/environment — inspect the folder structure
  2. GET /agent-api/v1/workspaces/:id/prompts — discover skills as callable prompts
  3. GET /agent-api/v1/workspaces/:id/file-tree — read the file tree
  4. POST /agent-api/v1/workspaces/:id/search — find indexed workspace context
  5. Do your agent's work using the files it can read
  6. POST /agent-api/v1/workspaces/:id/artifacts — write an output
  7. POST /agent-api/v1/workspaces/:id/files/:id/proposals — propose a memory update
  8. POST /agent-api/v1/workspaces/:id/events — report completion activity

6) Review your first proposal

When your agent proposes a change to memory/, you'll see a pending review in Filepad. Open it to inspect the diff line-by-line, then accept or reject.

That's the core loop: agent works, proposes changes, you approve what's right.

Next: Core concepts — the mental model behind the environment.