Puppyone
Git-native context drive for AI agents — file-level permissions, versioning, and multiple access methods.
Evidence shows: README claims file-level permissions (FLS) and audit logs, but no implementation details; SECURITY.md provides vulnerability reporting and key handling guidelines; Docker deployment defaults to sharing Docker socket, posing security risk; no user confirmation mechanism found; data flow is opaque; sensitive data handling only mentions environment variables; dependency security only mentions lockfiles and Dependabot; external effects include Docker socket mount and network services; rollback mentioned in README; source attribution not clear. Deductions: permissions and audit are only claims, no code evidence; user confirmation missing; data flow opaque; sensitive data handling insufficient; dependency security only mentioned; external effects risky; rollback unverified; source attribution unclear.
Evidence shows: README is consistent, but inconsistencies exist (e.g., SECURITY.md mentions PuppyAgent trademark, README mentions Puppyone); dependency availability unverified; failure messages not mentioned. Deductions: self-consistency deducted due to trademark inconsistency; dependency availability unverified; failure messages missing.
Evidence shows: README clearly identifies target users (AI agents) and scenarios (CLI, MCP, Git, SSH); capability boundaries partially described (e.g., supports multiple connectors); trigger precision not clear; environment fit good (supports Docker, cloud, self-hosted). Deductions: capability boundaries incomplete; trigger precision not clear.
Evidence shows: information architecture clear (README has TOC); install notes detailed (Docker, CLI); naming stability not clear; examples and FAQ limited; known limitations not clear; license is Apache-2.0; versioning/changelog missing; maintenance responsibility not clear. Deductions: naming stability not clear; examples and FAQ limited; known limitations not clear; versioning/changelog missing; maintenance responsibility not clear.
Evidence shows: output usability not clear; marginal value high (provides multiple access methods); cost-benefit not clear. Deductions: output usability not clear; cost-benefit not clear.
Evidence shows: claims in README partially traceable (e.g., feature list); cross-source corroboration limited (only README and SECURITY.md); fact-inference separation not clear. Deductions: cross-source corroboration limited; fact-inference separation not clear.
- Docker deployment defaults to sharing Docker socket, posing security risk; remote or multi-tenant deployments should use E2B sandbox.
- README claims file-level permissions and audit logs but lacks code evidence; further verification needed.
- SECURITY.md mentions PuppyAgent trademark, inconsistent with Puppyone in README, may affect brand consistency.
What does this agent do, and when should you use it?
Puppyone is a context hosting platform for AI agents, organizing data from various sources into a cloud file system called a Context Drive. Each agent gets a secure, isolated view of its files with Git-backed file-level permissions (FLS). The system supports multiple access methods — CLI, MCP, REST API, sandbox, and folder sync — letting agents retrieve context securely in the way they prefer. It offers self-hosted deployment via Docker Compose, bundling PostgreSQL, Supabase Auth, Redis, MinIO, and a FastAPI backend. The product aims to streamline AI agent adoption by centralizing and governing agent context, with provenance, collaboration tools, and APIs.
Puppyone runs a backend API (FastAPI) and a frontend (Next.js 15), orchestrating a Docker Compose stack (PostgreSQL, Supabase, Redis, MinIO). It ingests data from several SaaS sources (GitHub, Gmail, Google Drive, etc.), converting to Markdown/JSON and storing in the Context Drive. Admins create projects and define access points via CLI commands like puppyone access add url and puppyone access add mcp. Agents connect via MCP endpoints, REST API, or CLI using puppyone auth login and puppyone project use commands. The system enforces file-level security, runs agents in Docker or E2B sandboxes with only authorized files mounted, and provides audit logs, version history, rollback, and checkout/commit workflows.
- A coder using Cursor or Claude Desktop needs curated project context for code generation without messy retrieval — Puppyone delivers it via MCP.
- An AI agent team working across multiple projects wants to prevent cross-access — file-level permissions enforce security boundaries.
- An organization handling regulated data in a self-hosted setup needs audit trails and versioning for agent read/write operations.
- A developer wants to bring GitHub, Gmail, and Google Drive content together into agent workflows using built-in connectors.
- A builder of multi-agent systems needs different context sets for different agents, delivering one shared Context Drive with multiple scoped access points.
- A team using E2B sandboxes for secure code execution wants to run agents in isolation with only authorized files mounted.
What are this agent's strengths and limitations?
- Git-based file-level security (FLS) enforces per-agent access so unauthorized files are invisible.
- Multiple access points (CLI, MCP, REST API, sandbox, folder sync) provide integration flexibility.
- Built-in connectors for common data sources (GitHub, Gmail, Google Drive, etc.) reduce setup time.
- Full self-hosting via Docker Compose includes database, storage, and authentication out of the box.
- Requires multiple services (PostgreSQL, Redis, MinIO) making lightweight deployments heavy.
- Vendor dependencies: uses Supabase for auth and E2B as sandbox provider, potential lock-in.
- Requires Node.js (CLI) and Python (backend), multiple runtimes to manage.
- Documentation indicates OAuth connectors need extra configuration; self-hosting may be complex for some users.
How do you install or deploy this agent?
For cloud: sign up at https://www.puppyone.ai. For self-hosted: clone the repository and run cd puppyone/docker && cp .env.example .env && docker compose up -d. This starts the whole stack (frontend at http://localhost:3000) with one command. First startup takes 1-2 minutes. Optional: add ANTHROPIC_API_KEY to docker/.env to enable agent chat.
How do you use this agent?
Install the CLI: npm install -g puppyone. Log in: puppyone auth login (choose Cloud/Local/Custom URL). Create a project: puppyone project create "My Project" and puppyone project use "My Project". Add a webpage: puppyone access add url https://example.com --scope /refs. For MCP: puppyone access add mcp "My Context", then configure the output URL/API key in your MCP client. For self-hosted, log in with puppyone auth login -u http://localhost:9090.