Turn the same five steps into one button
Pick documents, search case law, extract a chronology, run a matrix, draft a memo, verify citations, export DOCX. Wire them once and run them on any matter. Start from a starter template, build your own, or type what you want and we generate it.
What You Get
Two ways to ship: starter templates or your own custom workflow
Eight published templates cover the most common litigator and transactional jobs out of the box. When the template does not fit, open the visual builder and chain any of 23 step types into your own pipeline. Both paths run on the same engine.
Natural-language workflow generation
Type "Draft a demand letter from the deposition summary and attach a chronology of events." The planner picks step types from the live registry, fills configs, runs the same five-gate validator a hand-built workflow uses, and lands you in the builder with a ready-to-run draft.
Versioned, publishable, run-on-any-matter
Every workflow has a draft and a published version. Publish snapshots the definition with a content hash so two runs of "v3" produce comparable artifacts. Launch a run against any matter; matter-scoped pickers thread through every step.
Branching, loops, and Test Run
Conditional branches route on a previous step's output. loop_over_documents fans out an inner step across every picked document in parallel. Test Run short-circuits side-effect steps (Save to Matter, Send Email) so you can sanity-check before a real run.
Validator that catches mistakes before you launch
Five gates check step types, configs, template references like {{step_2.documents[0]}}, branch targets, and nested loop configs. Unknown step type? Bad reference? Branch points upstream? You see a chip on the field before Test Run is enabled.
Live runs with per-step audit log
Watch a run tick through steps in real time. Each row shows inputs, outputs, latency, and any errors. Cooperative cancel via a single button. Full audit log of every step's inputs and outputs, exportable.
How Workflows Works in Vaquill AI
The 8 starter templates
Each template is a published, ready-to-run workflow you can launch against any matter. Fork a copy to customize without touching the original.
- ✓Quick Case-Law Memo: searches case law, drafts the memo, verifies citations, exports DOCX. A solid first workflow for research-heavy matters.
- ✓Demand-Letter Skeleton: searches supporting case law, drafts a demand letter, verifies citations, exports DOCX. Good first pass before lawyer review.
- ✓Document Triage: summarizes one matter document and exports the summary. Useful when a new file lands and you need a fast read.
- ✓Statute Cite Check: looks up a federal statute and verifies citations across an uploaded document.
- ✓Daily Citation Audit: searches a question across matter documents, summarizes findings, verifies any citations. A defensible scan for fabricated authorities.
- ✓NDA Bulk Review: waits for ingestion, runs a per-NDA matrix, then synthesizes outliers across the portfolio. Built for inbound diligence batches.
- ✓MSA Risk Score: scores an uploaded MSA against market-standard benchmarks (liability cap, IP indemnity, auto-renewal, etc.) with an overall risk tier.
- ✓Summarize Each Document: fans out Summarize across every picked document in parallel. Swap the inner step to score, audit, or extract.
23 step types across 8 categories
Every step is a typed building block with a defined input schema, output schema, and JSON-schema-rendered config form in the builder. Outputs feed inputs as references like {{step_2.documents[0]}}, written by the picker so you never touch the syntax.
- ✓Document I/O: pick_documents, save_to_matter, smart_auto_fill_inputs, wait_for_document_ingestion.
- ✓Retrieval: search_uploaded_documents, search_us_case_law, lookup_us_statute, web_search.
- ✓Extraction: summarize_document, extract_chronology.
- ✓Analysis: build_and_run_matrix, compare_two_documents, compliance_gap_check, score_against_benchmarks, cross_doc_synthesis.
- ✓Verification: verify_citations, synthesis_verifier.
- ✓Drafting: generate_legal_draft (jurisdiction-aware).
- ✓Routing: agentic_step, conditional_branch, loop_over_documents.
- ✓Output: export_docx_report, export_matrix_xlsx.
How natural-language generation actually works
The generator is not a freeform agent. It is a tightly-scoped pipeline that produces workflows the validator already knows how to check.
- ✓Classify: extracts intent, jurisdiction, and whether docs are needed from your prompt.
- ✓Retrieve: vector search over template descriptions and the live step registry, so the planner is grounded in real step types, not hallucinations.
- ✓Plan: LLM emits an ordered step list with rationale.
- ✓Slot-fill: LLM fills config per step, writing the right reference syntax behind the scenes.
- ✓Validate: the same five-gate validator that runs on hand-built workflows. Errors feed back into a repair loop with a hard retry cap.
- ✓Result: the builder opens with the workflow as a draft. Amber dots mark fields to confirm; rose dots mark fields still empty. Test Run is enabled when the validator passes.
Run loop, branching, and looping
A workflow is a DAG of typed steps. The executor walks it in order, threading outputs to inputs.
- ✓conditional_branch reads a previous step's output, evaluates an operator (equals, greater_than, contains, etc.), and routes to a true_branch or false_branch.
- ✓loop_over_documents takes a list and a configured inner step; it fans out the inner step across the list in parallel and reduces the results.
- ✓agentic_step opens a constrained tool-loop for the LLM with explicit tools, max-iterations, and an output schema. Useful when the next move depends on what the last step actually returned.
- ✓has_side_effects steps (Save to Matter, Send Email patterns) are short-circuited during Test Run so you can rehearse without writing anywhere.
- ✓Cooperative cancel via a Redis key: hit Cancel and the executor stops at the next safe boundary.
How It Works
Get started in minutes, not weeks.
Pick a starter, fork it, or start blank
Open Workflows in the sidebar. Pick one of 8 starter templates and fork to your org, or click New Workflow to start from a blank canvas.
Or just describe what you want
Click Generate, type "Review these vendor NDAs, flag off-market terms, and draft a summary memo," and Vaquill AI produces a draft workflow with steps, configs, and references already wired.
Edit in the visual builder
Add, remove, or reorder steps from the palette. Each step's config form is rendered from its JSON schema, with pickers for documents, matters, and upstream step outputs.
Validate and Test Run
The five-gate validator flags broken references, bad branch targets, and missing configs as chips on the field. Test Run executes the workflow with side-effect steps short-circuited.
Publish, then run on any matter
Publishing snapshots the definition as a versioned, content-hashed artifact. Launch a run against any matter; watch each step tick through the live run viewer.
Frequently Asked Questions
What is the difference between a template and a custom workflow?
A template is a published, system-authored workflow you can fork to your org. A custom workflow is one you built or generated. Both run on the same engine and use the same step types; templates are just a faster starting point.
Do I have to know any of this to use it?
No. Open Workflows, pick a template, run it on a matter. If you want something custom, type a sentence describing what you want and the natural-language generator builds a draft for you to review.
How is this different from Agent Mode?
Agent Mode is one-shot autonomous research that picks its own tools. Workflows are repeatable, named, versioned pipelines you author once and run many times. Agent Mode is for "I need an answer now"; Workflows are for "I run this every week."
What if my workflow references something that does not exist?
The five-gate validator catches it before Test Run is enabled. Unknown step type, broken reference like {{step_2.documents[0]}} pointing at a missing field, branch targeting an upstream step, or a nested loop config that does not type-check, all show up as chips on the offending field.
Can I branch or loop?
Yes. conditional_branch routes on a previous step's output. loop_over_documents fans out an inner step across every picked document in parallel. agentic_step opens a constrained tool-loop when the next move depends on what the last step returned.
What happens during Test Run?
The workflow executes end-to-end, but steps marked has_side_effects (Save to Matter, Send Email patterns) are short-circuited. You see what would happen without committing it to a matter.
Can I version workflows?
Every workflow has a draft and a published version. Publishing snapshots the definition with a content hash so two runs of "v3" produce comparable artifacts. The full version history is preserved.
Can I cancel a long-running workflow?
Yes. Cancellation is cooperative via a Redis key. Hit Cancel and the executor stops at the next safe boundary.
How is each step's work cited?
Steps that retrieve or read documents return source citations as part of their structured output. Downstream draft and verify steps thread those citations through, so the final DOCX has inline citations you can verify.
Will my workflows work with team members?
Workflows are org-scoped. Anyone in your organization with the right role can list, fork, edit, and run org workflows. Templates are org-wide too.
Stop redoing the same five steps for every matter
Build one workflow, run it on every matter. Or pick a starter and run it now.