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 Modulo.

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. Modulo 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.

Modulo

Modulo is self-hosted by design, with no hosted SaaS edition offered today (a read-only demo instance is live at https://app.modulo.run/demo). It runs on Docker Compose or Fly.io under the Business Source License (BUSL) with a commercial/Enterprise 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 immutable SHA-256 linked audit chain, bounded autonomy with a level you set per pipeline and eval evidence for raising it, and SOC 2 evidence export. The pipeline model is built around SDLC concepts - PR triggers, code review gates, deployment approvals - and dispatches any agent via sandbox, across GitHub, GitLab, any model, and any agent.

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 Modulo - 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.

Modulo 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 Modulo vs Factory.ai, Modulo vs n8n, and Modulo vs Dify.

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