What is Runsight?
Runsight is a YAML-first workflow engine built specifically for AI agents. It lets developers design, commit, run, and evaluate agent workflows with Git-native version control. Every workflow is stored as a plain YAML file in your repository, so you can branch, review, and merge agent changes just like any other code. The platform runs entirely on your machine with your own API keys, is 100% open source under the Apache 2.0 license, and requires no cloud account or signup to get started.
Who it's for
- AI/ML engineers who need to version-control agent workflows and track execution costs to the cent
- Developer teams that want a shared, reviewable format for agent orchestration instead of scattered Python scripts
- Budget-conscious builders who need hard spending caps and real-time cost visibility to prevent surprise bills
Key features
Canvas and YAML editor in sync
Two views of the same workflow state. Edit YAML in the Monaco editor and the canvas updates automatically. Move nodes on the canvas and the YAML stays clean. Commit the YAML to Git like any other code, with full diff support for code reviews.
Per-run cost tracking with hard budget caps
Every block and every run is tracked to the cent. Set a max_cost and max_tokens budget that kills execution before overspend occurs. No more surprise bills after batch runs — you see exactly what each step costs in real time.
Built-in evaluation framework
Assertions on every block output let you validate results programmatically. Transform hooks enable structured extraction, and regression testing across runs ensures your agents behave consistently as workflows evolve.
Pause and kill mid-run
Stop a running agent before it wastes your budget. Pause execution, inspect the current state, then either resume or kill the run. This gives you fine-grained control over long-running or expensive agent tasks.
What stands out
"Ship agents like you ship code."
This philosophy is the core of Runsight. By treating workflows as Git-diffable YAML files, the platform brings software engineering best practices — branching, code review, version history — into the world of AI agent development. If Runsight disappears tomorrow, you still have readable, executable workflow configs. No vendor lock-in, no proprietary formats, no cloud dependency.
Worth checking out if…
You're building AI agent workflows and want version control, cost transparency, and evaluation built into your toolchain from day one. Runsight is especially valuable if you're tired of debugging agents with print statements, managing scattered Python orchestration scripts, or getting surprised by API costs after batch runs. The one-command startup (uvx runsight) and self-hosted nature make it trivial to try — no signup, no cloud account, just your machine and your API keys.
