Reviewed by Harsh Desai · Last reviewed:
Runsight
A YAML workflow builder that lets you design, version and cost-control AI agents directly in Git.
Best for
What does Runsight do?
- •YAML in Git define entire agent workflows as version-controlled YAML files stored directly in your repository.
- •Dual editors edit with a visual canvas and Monaco YAML editor that sync bidirectionally in real time.
- •Cost tracking monitor exact per-block and per-run LLM costs with enforceable hard budget caps.
- •Assertions framework add built-in assertions to validate outputs and catch errors automatically.
- •Transform hooks insert custom transform hooks at any step to modify data on the fly.
- •Regression testing run a full regression testing framework to compare agent behavior across versions.
- •Pause and resume pause agents mid-run, inspect current state, then resume or kill execution safely.
- •Loop support build loops with explicit break conditions and create nested sub-workflows.
- •One-command start launch locally using uvx runsight with no signup, no account, and no cloud dependency.
- •Self-hosted models run everything on your own models and API keys with full data control.
- •Multi-platform works on Linux, macOS, Windows and Web using your existing infrastructure.
- •Budget Caps Enforce hard per-run spending limits using tracked costs across all LLM calls in your YAML workflows.
- •State Inspection Pause any agent mid-execution to examine full internal state before choosing to resume or kill the run.
- •Nested Workflows Compose complex agents by calling nested sub-workflows directly from within parent YAML files with full state passing.
- •Bidirectional Sync Modify workflows in real-time using the canvas view or Monaco editor with automatic two-way updates between both interfaces.
Pricing:
- •Free open source $0/mo: completely free with full source code and no usage limits.
- •Self-hosted only $0/mo: no paid cloud tiers or hosted execution available at any price.
- •Local forever $0/mo: run on your hardware with your keys and zero vendor costs.
What are Runsight's limitations?
- •YAML required demands solid familiarity with YAML syntax before building complex workflows.
- •Self-host only you must manage your own infrastructure with no managed hosting option.
- •No cloud runs lacks any hosted or cloud execution environment for quick testing.
- •Early stage still evolving so some advanced features remain incomplete or experimental.
Our Verdict
For the Vibe Builder, Runsight transforms YAML workflows into a canvas for crafting Git-native AI agents that feel alive and intentional. Its declarative syntax lets creators sketch agent behaviors like poetic blueprints, where each step in the pipeline resonates with purpose and flows naturally from one creative impulse to the next. The open-source foundation encourages endless remixing inside your own repositories, turning infrastructure into an extension of your aesthetic rather than a barrier. This approach rewards those who see code as atmosphere, delivering a lightweight yet expressive way to orchestrate intelligent systems without leaving the comfort of version control.
For the Developer, Runsight offers a pragmatic, Git-centric framework for building reliable AI agents through familiar YAML configuration that integrates directly into existing CI/CD practices. Because it is completely free and open source with no paid tiers or cloud hosting, teams retain full control over data and execution while embedding agent logic alongside application code. The requirement to self-host and manage infrastructure locally ensures security and customization but demands solid DevOps comfort. Its evolving feature set rewards iterative development, allowing developers to version, review, and collaborate on agent behaviors exactly as they would any other codebase.
Honest limitation: Runsight remains early-stage with an evolving feature set and requires familiarity with YAML syntax for workflow design, which can slow onboarding for those preferring visual builders or higher-level abstractions. It must self-host and manage infrastructure locally with no hosted or cloud execution option available, increasing operational overhead for teams seeking instant scalability. These constraints make it less suitable for rapid prototyping or organizations without dedicated infrastructure expertise. Overall it earns a 7.4/10 for teams already embedded in Git-centric workflows but falls short for those needing polished, zero-ops experiences.
Skip it if you want managed execution and visual tooling; consider LangGraph instead for more mature orchestration abstractions.
Related Tools
View all
Compare Runsight With
Also Useful For
Frequently Asked Questions
What is Runsight and how does it define AI agent workflows?
Runsight is an open source framework that lets you build AI agent workflows by defining tasks, tools, and decision paths in simple YAML files that the engine then executes locally. It defines these workflows as a series of connected nodes where each node represents an LLM call, tool invocation, or conditional branch, making complex agent behaviors reproducible without writing boilerplate code. You run everything on your own hardware so data never leaves your machine.
Is Runsight free to use in 2026?
Yes, Runsight is free to use in 2026 because it operates on a completely free open source model with no usage limits or paid tiers. You download the full source code and run it locally forever without any vendor costs. There is a free tier available for everyone.
Who should use Runsight for agent building?
Developers and technical teams who want full control over their AI agents should use Runsight for agent building since it runs locally on your hardware with your own API keys. It suits users who prefer defining workflows in YAML and need zero ongoing vendor dependency. Non-technical users or those wanting a hosted platform will likely find it too low-level.
How does Runsight pricing compare to CrewAI?
Runsight pricing is free open source at $0/mo, self-hosted only at $0/mo, and local forever at $0/mo, while CrewAI follows a similar open source approach but some users report needing paid LLM services on top. Runsight has no paid cloud tiers or hosted execution available at any price, so total cost stays at zero if you already have hardware and keys. This makes it cheaper for teams avoiding any vendor lock-in compared to CrewAI setups that sometimes scale into usage-based fees.
What alternative exists if I dislike YAML syntax?
If you dislike YAML syntax the main alternative is to use Python-based agent frameworks like LangGraph or AutoGen that let you define workflows directly in code instead of configuration files. Runsight itself stays YAML-first, so switching tools is the practical route for code-centric builders.
Affiliate link: we may earn a commission. How this works.
Runsight
Free tier available