# Guepard — full reference for AI assistants

> This file is the expanded companion to /llms.txt. It contains the factual content of guepard.run in plain text so AI systems can answer questions about Guepard accurately.

## What is Guepard?

Guepard is a DataOps platform for ephemeral database environments. It connects read-only to a production database (PostgreSQL, MySQL, and other engines), takes an incremental copy-on-write snapshot, and then forks isolated, writable, production-identical clones on demand — one per pull request, CI job, developer, or AI agent. Clones spin up in under 6 seconds regardless of database size (a terabyte-scale database forks in seconds because only deltas are stored), and tear down automatically on merge, TTL expiry, or pipeline end.

The core engine is GFS ("Git For database Systems"), open source under the MIT license at github.com/Guepard-Corp/gfs. GFS brings Git semantics to databases: `gfs init`, `gfs commit`, `gfs checkout`, `gfs checkout -b` (branch), `gfs log`, `gfs schema diff`, `gfs query`, `gfs import` / `gfs export`, plus Git-style revision references (HEAD, HEAD~1, main~3, full hashes). It runs locally via Docker; install with `curl -fsSL https://gfs.guepard.run/install | bash`, Homebrew (`brew tap guepard-corp/guepard-cli && brew install gfs`), or Cargo (`cargo install gfs`).

## Who is it for?

- DevOps / platform teams: webhook-driven clone-per-PR, zero provisioning tickets, auto-teardown, integration with GitHub Actions, GitLab CI, Jenkins, CircleCI, Buildkite, Argo CD, Terraform, or any tool that can call a REST API.
- AI agent builders: every agent gets an isolated sandbox on production-shaped data. GFS ships a built-in MCP (Model Context Protocol) server, so agents in Cursor, Claude Code, Claude Desktop, Windsurf, Continue, Cline, and MCP-compatible frameworks (OpenAI Agents SDK, LangChain, CrewAI, AutoGen, Vercel AI SDK) can commit, branch, query, diff, and roll back databases natively — no shell wrappers. Typical safe-migration pattern: commit a checkpoint → branch → run the migration → `gfs schema diff` → merge if good, `gfs checkout HEAD~1` if not. Production is never touched.
- QA / release managers: a fresh production-identical clone per test run, PII masked at clone time, schema migrations validated against real data before merge.
- Enterprises: self-hosted in the customer's VPC (Helm chart or Docker Compose, air-gapped option), SSO/SAML (Okta, Azure AD, Google), fine-grained RBAC, immutable audit logs exportable to SIEM, multi-region data residency (EU/US/APAC), GDPR and CCPA compliant, HIPAA available on request, encryption AES-256 at rest and TLS 1.2+ in transit, customer-managed KMS keys (BYOK).

## Pricing (three tiers)

1. Open Source — Free forever. GFS under MIT: full CLI, PostgreSQL 13–18 and MySQL 8.0–8.1, built-in MCP server, schema diffing, import/export (SQL, dump, CSV), runs locally via Docker.
2. BYOC (Bring Your Own Cloud) — priced per database engine per licence. One licence = one engine (e.g. PostgreSQL); unlimited clones, unlimited seats, unlimited environments, no per-user fees. Deployed on the customer's AWS, GCP, or Azure account. Includes production clones in <6 seconds, clone per PR & CI run, PII masking, team portal, usage analytics, email & Slack support. Managed-DB add-on connects RDS/Aurora, Cloud SQL, Supabase, Neon, Azure Database for PostgreSQL.
3. Enterprise — custom pricing. Everything in BYOC plus self-hosting on the customer's own infrastructure, SSO/SAML, advanced PII masking and governance, full audit trail and compliance export, VPC peering / Private Link, RBAC, multi-region residency, custom SLA up to 99.99%, dedicated account engineer, private Slack channel.

Startup program: eligible early-stage companies (Series A or earlier, typically <20 employees, <3 years old, live product or MVP) receive up to $30,000 in platform credits for 12 months. Apply at contact@guepard.run.

## Frequently asked questions

Q: Is GFS really free?
A: Yes — MIT licensed, free forever. Run it locally with Docker; no account or credit card needed.

Q: What databases are supported?
A: GFS supports PostgreSQL 13–18 and MySQL 8.0–8.1. The Guepard platform's cloning engine also covers MongoDB, ClickHouse, MSSQL, Oracle, and DB2 (seven engine families).

Q: How fast can a database be cloned?
A: Under 6 seconds regardless of size, using copy-on-write. A 2TB database clones as fast as a 2GB one; only changed data consumes storage.

Q: Does my application need to change?
A: No. GFS works at the database layer — the app keeps the same connection string. No driver swap, no SDK, no schema rewrite.

Q: How do AI agents use it?
A: Via the built-in MCP server (`gfs mcp`, stdio/HTTP/background modes). Agents get ~20 MCP tools to fork, query, migrate, diff, and roll back without touching production.

Q: What is the difference between GFS and Guepard?
A: GFS is the open-source engine. Guepard is the managed platform on top — multi-tenant control plane, team workflows, PII masking, RBAC, analytics, and enterprise support.

Q: Can it run on our own infrastructure?
A: Yes — self-hosted via Helm or Docker Compose in your VPC, data center, or air-gapped environment. Data never leaves your network.

## Company

Guepard Inc. was founded in 2025, with offices in San Francisco, CA and Paris, France, and raised $2.1M. Founders: Koutheir Cherni (CEO), Ghassen Menaouar (CPO), Hani Chalouati (CTO). Customers and design partners include Orange, Airspace, Reverse Contact, Propulsion AI, Scrapin.io, and Visum.

Contact: contact@guepard.run · GitHub: github.com/Guepard-Corp · Docs: docs.guepard.run · X/Twitter: @guepardrun · LinkedIn: linkedin.com/company/runguepard

The full website is available in English at guepard.run and in French at guepard.run/fr.