Self-hosting is a key differentiator in the agent platform market. For teams in regulated industries, air-gapped environments, or anyone who wants full control over their data, it is the only option that works for many teams. Three platforms are built around it: Tembo (Agent Studio), Dify, and Farna Control.

Two more ways to run inside your own boundary emerged while this post was being written. GitLab Self-Managed carries GitLab's governed software factory - announced 6 October 2026 - onto your own servers, and AWS Bedrock AgentCore runs agents under their own AWS identity inside your AWS account. The question that separates them is not whether self-hosting exists, it is how you obtain it. GitLab Self-Managed and AgentCore are self-host available and sales-gated: negotiated through an enterprise conversation, at a price you discover through sales. Farna Control is self-host at every tier with no sales gate, including the free Community edition, bought without talking to anyone.

Tembo Agent Studio

Tembo's Agent Studio (launched May 2026) is MIT-licensed and self-hosted via Docker Compose, from source or published GHCR images. Agent definitions live in a Git repository as the source of truth, so every change to an agent produces a pull request - the same reviewable artifact teams already trust - and it orchestrates Claude Code, Codex, and Cursor's background agents. It is agent- and model-agnostic, with Git-native config review, and audit logs and identity stay inside your environment. Version 2026.9.1 (September 2026) added optional per-agent eval suites whose assertions must pass before a draft can be promoted, so agent quality is gated at authoring time; there is still no documented cross-run evidence record or progressive-autonomy model (see tembo.io), and its focus is agent coordination plumbing rather than SDLC.

Best for: Teams who want agent-agnostic orchestration with Git-native config and are comfortable managing infrastructure.

Dify

Dify is the most established self-hosted AI application platform. Its community edition offers visual workflow building, RAG pipelines, model provider integration, and a large plugin ecosystem, and it is model-agnostic: any model via API. It is licensed Apache 2.0 or commercial, deployed via Docker Compose, and horizontal: it can be used to build SDLC workflows but is not designed for them. Governance is basic RBAC and audit in the Enterprise tier.

Best for: Teams building general AI applications who want a self-hosted visual workflow builder.

Farna Control

Farna Control is an Agentic Delivery Governance platform for the software lifecycle, self-hosted by design, with no hosted SaaS edition offered today (a read-only demo instance is live at https://app.farna.dev/demo). It runs on Docker Compose or Fly.io under the Business Source License (BUSL) with a commercial option - source-available for self-hosting, not a traditional open-source license - with no telemetry by default. The entire system lives on your infrastructure: pipeline definitions, agent configurations, audit logs and gate configurations. Governance is first-class: deterministic gates, HITL review, an append-only SHA-256 linked audit chain, bounded autonomy with a level you set per pipeline and eval evidence for raising it, and audit export. The pipeline model covers SDLC concepts such as PR triggers, code review gates and deployment approvals, while dispatching any agent that meets its contract via sandbox across GitHub and GitLab, with your choice of supported model backends.

Best for: Teams who want governed, self-hosted agent automation for their SDLC specifically.

The trade-offs

Tembo's Agent Studio is the closest structural match to Farna Control - self-hosted, git-native, agent-agnostic. Its governance covers how agent definitions are versioned and reviewed, not a compounding evidence record for what an agent may do autonomously over time, so choose it for Git-native agent configuration management.

Dify offers the broadest platform but the shallowest SDLC depth, a good choice when you need a general AI application platform and SDLC governance is secondary.

Farna Control is the most opinionated of the three, built specifically for SDLC governance, so it does less outside that domain.

For a deeper look at what self-hosted agent governance requires, see the self-hosted agent governance guide. For head-to-head comparisons, see Farna Control vs Factory.ai, Farna Control vs n8n, and Farna Control vs Dify.

The principle behind this: self-hosted, no telemetry - your infra, your data, no cloud dependency.