Open Source

YAML workflows.
Git-native AI agents.

Design agent workflows in YAML. Commit to Git. Track cost per run. Evaluate with built-in assertions. Pause or kill agents mid-flight.

$ uvx runsight View on GitHub
canvas
▷ Research $0.03
LINEAR
◆ Review $0.02
GATE
◼ Publish
CODE
workflow.yaml 
version: "1.0"
blocks:
  research:
    type: linear
    soul_ref: analyst
  review:
    type: gate
    soul_ref: editor
    eval_key: research
    pass: publish
    fail: research
  publish:
    type: code

workflow:
  name: Research Pipeline
  entry: research
YAML Filesystem-native workflows
Git Every save is a commit
$0.00 Per-run cost tracking
100% Open source
THE DUALITY

Canvas and YAML. Same state.

Two views of the same workflow. Edit YAML in the Monaco editor — the canvas updates. Move nodes on the canvas — the YAML stays clean. Commit the YAML to Git like any other code.

terminal 
# Your workflow is a file on disk
$ cat custom/workflows/research-pipeline.yaml

version: "1.0"
blocks:
  research:
    type: linear
    soul_ref: analyst

# Git diff it like any other code
$ git diff workflow.yaml
- soul_ref: analyst
+ soul_ref: senior-analyst
+ budget:
+   max_cost: 0.50

# Commit and run
$ git add . && git commit -m "upgrade analyst soul, add cost cap"
$ runsight run research-pipeline
CAPABILITIES

Ship agents like you ship code.

Per-run cost tracking

Every block, every run — tracked to the cent. Set hard budget caps that kill execution before overspend. No more surprise bills.

budget: max_cost: 2.00 max_tokens: 50000 # Run output ▶ research $0.032 1.2s ▶ review $0.018 0.8s ───────────────────── Total: $0.050 Cap: $2.00

Built-in eval framework

Assertions on every block output. Transform hooks for structured extraction. Regression testing across runs.

Pause and kill mid-run

Stop a running agent before it wastes your budget. Pause execution, inspect state, then resume or kill.

Git-diffable workflows

Every workflow is a YAML file in your repo. Branch, review, and merge agent changes like any other code.

Self-hosted. Your keys.

Runs on your machine. Your API keys. Your models. Nothing leaves your network unless you want it to.

Loops and sub-workflows

Iterative refinement with break conditions. Nest workflows as child steps. Hierarchical orchestration out of the box.

THE DIFFERENCE

Stop debugging agents in the dark.

Without Runsight
  • Agent workflows scattered across Python files
  • No visibility into running agent decisions
  • Debugging is printf and prayer
  • Cost surprises after every batch run
  • Three teams, three approaches, no shared patterns
  • No eval — "looks good" is the test
With Runsight
  • Every workflow is a YAML file in Git
  • Every block traced with cost and latency
  • Visual canvas + YAML editor, same state
  • Hard budget caps — kill before overspend
  • One tool, one format, one Git history
  • Assertions and regressions on every run
HOW IT WORKS

One command. Open your browser.

01

Run

One command. No signup. No cloud account. Scaffolds a project if none exists.

$ uvx runsight
# → localhost:8000
02

Design

Write YAML in the editor or drag blocks on the canvas. Add your API key in the onboarding flow.

# Your workflow is a file on disk
custom/workflows/my-flow.yaml
03

Execute

Click Run. Watch blocks execute in real time. See cost accumulate per step.

▶ research $0.03 1.2s
▶ review $0.02 0.8s

YAML workflows. Git-native.
Open source.

Your workflows are YAML files in Git. If Runsight disappears, you still have readable config. No vendor lock-in by design.

$ uvx runsight View on GitHub Read the docs