AI Workflow Builder
A real working multi-step LLM workflow tool. Three curated presets ship in v0.1 (Research → Summarise → Tweet · Code → Review → Refactor · Email → Classify → Reply) but every prompt and input is editable in place. Each step streams independently; outputs flow into the next via {{stepId.output}} templating. Sits alongside the existing AI ops toolkit: ai-eval evaluates single calls, PR Reviewer is a fixed 4+1 workflow, this is the user-defined version. Lives at /workflow.
Frontend
Backend
Database
Tools
- • Most prompt-chain tools require visual flow editors that get unwieldy fast; raw YAML editors lose the wow factor
- • Showing real-time step execution requires per-step streaming, error isolation, and visible state
- • Outputs flowing into prompts is fiddly — needs template substitution that's obvious in the UI
- • Preset-first UX: pick a workflow, see all steps + inputs immediately, run with one click; advanced users edit any prompt inline
- • Per-step state machine (idle → running → done/error) with per-step latency surfaced; final-output card highlights the deliverable
- • Simple {{stepId.output}} and {{input.key}} substitution shown plainly in the prompt-template view — no hidden magic
Working tool with three useful real-world chains out of the box
Built on the existing /api/demo/generate route — no new server plumbing required
Natural sibling of ai-eval (evaluate single steps) and PR Reviewer (fixed workflow)
Other Projects
CLI + web viewer that runs your prompts against test cases with contains / regex / equals / llm-judge assertions, produces a JSON report, and fails the build on regressions in CI.
View ProjectPaste any public GitHub PR — four specialised AI agents review it in parallel (correctness, security, style, tests), a lead reviewer synthesises a severity-graded verdict.
View Project