Question 1

What is PuppyOne?

Accepted Answer

PuppyOne is a file-based workspace built for AI agents. It ingests your SaaS data (Notion, Slack, Gmail, Postgres, and more) into a unified file system, lets multiple agents collaborate with granular per-agent permissions, and tracks every change with Git-style version control. Think of it as the shared, versioned hard drive your AI pipeline never had.

Question 2

Why do AI agents need PuppyOne?

Accepted Answer

Three reasons. First, agents generate valuable outputs — research, code, reports — but most runtimes lose everything when the session ends. PuppyOne backs up every piece of context the moment it's written, so nothing is ever lost. Second, when you run multiple agents (Claude for research, GPT for drafting, a custom agent for data processing), they need to share files without stepping on each other. PuppyOne gives each agent scoped permissions in one shared workspace. Third, your context is scattered across Notion, Slack, Gmail, databases — PuppyOne pulls it all into a single file-based workspace agents can read with cat, grep, and ls. Backup, collaboration, aggregation — that's why agents need PuppyOne.

Question 3

My agent loses all its output when the session ends or crashes. How does PuppyOne prevent data loss?

Accepted Answer

This is exactly what PuppyOne is built for. Every file an agent writes is immediately committed to a versioned workspace — not just held in session memory. When the session ends or the agent crashes, every output is still there, intact, with a full diff of what changed. No more re-running pipelines to recover lost work. You can also checkpoint before risky operations and roll back to any previous state in milliseconds.

Question 4

Which agents and tools can connect to PuppyOne? (n8n, OpenAI, Claude, Cursor, Manus…)

Accepted Answer

If a tool can send an HTTP request or speak MCP, it works with PuppyOne. Confirmed integrations include: n8n (via HTTP Request node), OpenAI Assistants API and Agents SDK, Claude / Anthropic via native MCP, Cursor via MCP, LangChain and LangGraph, CrewAI, AutoGen (Microsoft), and Manus. Each integration gets its own Access Point with scoped permissions — so multiple tools can safely share the same workspace without interfering with each other.

Question 5

How do multiple agents work together without overwriting each other?

Accepted Answer

Each agent gets an Access Point — a scoped entry into the workspace with explicit read/write paths. Agent A might have write access to /finance and read-only to /shared, while Agent B owns /docs. Conflicts are prevented by path isolation; for coordinated writes, agents can work on separate branches and merge via the API when ready.

Question 6

Can I give different permission levels to different agents?

Accepted Answer

Yes — this is core to PuppyOne. Every Access Point defines which paths an agent can read, which it can write, and which are completely hidden. An untrusted external agent can be given read-only access to a single folder. An internal orchestrator can have full workspace access. Changes to permissions take effect immediately without redeployment.

Question 7

What SaaS data sources can I sync into PuppyOne?

Accepted Answer

Current connectors include Notion, Slack, Gmail, Google Sheets, Postgres, MongoDB, Linear, Jira, and Airtable, with more being added continuously. Data is synced, cleaned, and written as structured files (.json, .md, folders) that any agent can read via Bash, MCP, or API. You can also push arbitrary data via the API if your source isn't listed yet.

Question 8

How does version control work for agent-generated files?

Accepted Answer

Every write operation creates a commit with the agent identity, timestamp, and a diff. You can view the full history with a git log-style API, branch the workspace before a risky operation, and merge branches when done. The version graph is queryable — you can ask 'what did agent-b change in /finance between 2pm and 4pm yesterday' and get a precise answer.

Question 9

What happens if an agent deletes or corrupts a file?

Accepted Answer

Every state of the workspace is preserved. Open the audit log, find the last known-good commit, and call the rollback API with that commit hash. The workspace is restored to that exact state in milliseconds. You can also roll back a single file without affecting the rest of the workspace.

Question 10

How do agents actually access the workspace?

Accepted Answer

Three access methods: (1) Bash / Shell — agents running in a container can cat, grep, ls, and write files as if it were a normal filesystem. (2) MCP — a native Model Context Protocol server for Claude, Cursor, and any MCP-compatible client. (3) REST API — language-agnostic, works with any HTTP client. All three methods share the same permission model and version history.

Question 11

How is PuppyOne different from just giving agents a shared folder or Dropbox?

Accepted Answer

A shared folder has no concept of agent identity, permissions per agent, or structured version history. Dropbox has no Bash access, no MCP support, no SaaS ingestion pipeline, and no branch/merge semantics. PuppyOne is built specifically for multi-agent pipelines: every operation is attributed to an agent, every path can be permission-scoped per agent, and the entire workspace is Git-versioned from day one.

Question 12

Can I self-host PuppyOne? Does data leave my VPC?

Accepted Answer

Yes. We offer a self-hosted Docker deployment for enterprise environments. In self-hosted mode, all file storage, metadata, and version history stay entirely within your infrastructure — nothing is sent to our servers. You can optionally connect the control plane for monitoring and updates, but this is not required. Full data sovereignty, zero lock-in.

Question 13

How can I get support or contact the team?

Accepted Answer

Email the founder directly at guantum@puppyone.ai, DM us on Twitter at https://x.com/puppyone_ai, or join our Discord community for real-time support and to share what you're building. We read everything.

Frequently Asked Questions