filepadai
sign inget started

The workspace
AI work depends on.

A structured, visible, versioned environment where AI agents read context, work with files, and propose changes humans can review.

filepad.ai/w/my-workspacefilesdiffagents
.filepad/
agents/
research-agent.md
skills/
memory/
facts.md
sources/
artifacts/
agents/research-agent.md
1# Research Agent Notes
3name: external-agent
4model: gpt-4o
6## Scope
7read:
8 - agents/, skills/, memory/
9write:
10 - artifacts/
11propose:
12 - memory/
memory/facts.mdpending
···## Operational Facts
primary_language: Python
+primary_language: TypeScript
···
+module_pattern: canonical
✓ Accept✕ Reject
external-agent · 2h ago
● external-agent connected1 pending reviewmy-workspace

AI agents are powerful, but hard to trust and manage.

Not because they can't do useful work. Because you can't always see, control, or verify what they're using.

01

Context is scattered

Instructions, prompts, and source material live across repos, docs, and hidden dashboards. There’s no single place to understand what your agent actually depends on.

02

Knowledge is hard to inspect

Important context sits in vector stores, databases, and past conversations. You can't easily read it, correct it, or know if it's still right.

03

Changes are hard to trust

When an agent updates something, you don't know what changed or why. No diff. No audit trail. No review.

What should be a system becomes fragments.

One system for the context AI work depends on.

Filepad gives instructions, memory, sources, uploads, outputs, and automations a real place to live — as visible folders, editable files, version history, and changes humans can review and approve before they become part of the workspace.

.filepad/

Managed workspace metadata.

agents/

External-agent notes and operating context.

skills/

Reusable instructions and procedures.

memory/

Durable facts, decisions, and working context.

sources/

Grounding material: documents, data, references.

uploads/

Raw inputs before they become usable.

artifacts/

Outputs from humans, FilepadAI, or external agents.

automations/

Reviewed automation definitions and runs.

filepad.ai/w/my-workspace /research-agent.md— real folders, real files, visible before AI acts on them

Four flows. One coherent system.

01Build

Build the workspace

Add source documents. Define reusable skills. Curate memory. You are building visible context before AI touches anything.

skills/ memory/ sources/
02Connect

Connect an agent

Give an external agent scoped API access. Define what it can read, what it can write, and what requires your approval. Permissioned from day one.

read sources/ write artifacts/ propose memory/
03Operate

Operate with review

Agents work against the file tree. Risky writes become proposals. You review diffs, accept or reject changes, and always know what happened.

primary_language: Python primary_language: TypeScript
04Maintain

Maintain the workspace

FilepadAI helps keep the workspace healthy: improve instructions, create skills, explain what an external agent can see, and debug bad behavior.

FilepadAI › external-agent has read access to memory/ and write access to artifacts/. There is 1 pending review.

A real interface for external agents.

Filepad now exposes a published SDK and MCP server so outside agents can read workspace context, search indexed files, create artifacts, propose edits, and report what happened without scraping the UI or touching private routes.

Open source

SDK and MCP server

The public Agent Access SDK and MCP server are MIT-licensed and live under the Filepad GitHub org.

github.com/filepad/agent-access
Scoped

Keys grant only what an agent needs

Read the workspace, create artifacts, propose edits, write events, or create signals. Each capability is explicit.

env:read · artifacts:write · files:propose
Reviewed

File edits come back as proposals

External agents can suggest changes without silently mutating Filepad-managed files. Humans approve the diff.

propose → review → approve
Portable

Works with MCP-capable agents

Use the same Filepad workspace from OpenClaw, Claude Desktop, Cursor, Windsurf, or a custom runtime.

npx -y @filepad/mcp-server@latest

Built for people who operate
agents, not just build them.

01
AI consultants

Build and hand off AI-ready workspaces to clients with full visibility into what was created.

02
Founders

Wire agents into core workflows without losing track of what they know and do.

03
Agencies

Manage AI-ready workspaces across multiple clients from one structured surface.

04
Internal AI teams

Standardize how agents are configured, reviewed, and maintained across the organization.

05
Developers

An inspectable, API-accessible context layer you can point any agent at and trust.

8
standard environment folders
0
hidden databases your agents read from
100%
of agent context visible and editable
agents you can connect, all scoped

Simple. No usage-based surprises.

Start free. Scale when you need more agents or workspaces.

starter
$0/month

For individuals exploring agent workflows.

  • 1 workspace
  • 1 connected agent
  • Full file editor + diff review
  • 100 FilepadAI messages / mo
  • No audit log
  • No multi-agent
Get started
pro
$49/month

For operators running real agent workflows.

  • 3 workspaces
  • 5 connected agents
  • Full file editor + diff review
  • Unlimited FilepadAI messages
  • Full audit log
  • API access
Start free trial
team
$149/month

For agencies and teams managing multiple environments.

  • Unlimited workspaces
  • Unlimited agents
  • All Pro features
  • Team members + roles
  • SSO + priority support
  • Custom retention policies
Contact us

Your agent's operational brain.

Finally visible.

Start with a free workspace. Connect your first agent in minutes. No configuration hidden from you, ever.

Free to start · No credit card required · Cancel anytime