Digesto — How to Use

Last Updated: July 2026

Requirements

  • Windows 10 or Windows 11 (64-bit) with .NET 8 desktop runtime
  • One of the following AI providers (choose in Settings):
    • Anthropic (Claude) — requires an Anthropic API key and internet access
    • OpenAI (GPT) — requires an OpenAI API key and internet access
    • Ollama (local) — no API key or internet required; Ollama must be installed and running locally
  • A GitHub account and a fine-grained Personal Access Token scoped to your rule-packs repository (for git sync)
  • Internet access for cloud AI ingestion and git sync — Browse Rules works offline once packs are cloned; Ollama ingestion works fully offline

Digesto is a standalone executable — it does not require Revit. Run AvantLeap.Tools.Digesto.exe directly; there is no Revit ribbon or Add-In Manager entry.

First-time Setup

Open Digesto and click Settings. Configure:

  1. Git identity — the name and email used for rule-pack commits.
  2. Rule-packs repository — the GitHub repository URL where rule packs are stored, and your fine-grained Personal Access Token (least-privilege: scoped to only that repository).
  3. AI provider — choose your preferred provider:
    • Claude or OpenAI: paste the matching API key.
    • Ollama (local): enter the base URL (default http://localhost:11434) and the model name (e.g. llama3.1). No API key is required.

API keys and the GitHub token are encrypted at rest (Windows DPAPI, CurrentUser scope) inside %APPDATA%\AvantLeap\Digesto\settings.json. They are never committed to a repository, never written to log lines, and never placed in a git remote URL. DPAPI ties the encryption to the current Windows user — copying settings.json to another machine yields empty secrets.

The Launcher

Digesto opens a five-button launcher. Each button opens a dialog:

Button What it does
Ingest Document Load a regulatory document → AI extraction → review → validate → save & commit
Browse Rules Explore stored rule packs by domain and keyword
Compliance Graph Visualize which rule packs feed which consuming tools
Settings Git identity, rule-packs repo, AI keys, Ollama config
Rule & Ecosystem Map Live diagram of document types → compliance patterns → gatekeepers and providers across the AvantLeap ecosystem

Ingest Document

This is the core workflow — it converts a regulatory document into rules.

  1. Click Ingest Document.
  2. Select a source file — PDF, plain-text extract (.txt), or structured draft (.md).
  3. Digesto extracts the document text. For cloud providers, this text is sent to your chosen AI API; for Ollama, it is processed entirely on your machine.
  4. The AI returns structured draft rules based on the versioned extraction prompt.
  5. Review the extracted rules in the dialog. AI-extracted content is a draft — review every rule for accuracy before promoting.
  6. Validate — rules are checked against the canonical schema (schema/rule-schema-v1.json) on save. Invalid rules are flagged and must be corrected before saving.
  7. Save & commit — validated rules are written to the rule-packs repository as JSON (and accompanying .md) and committed to git.

Draft → validated gate: all AI-extracted rules start as draft. A human reviewer must promote them to validated before downstream compliance tools should rely on them. This gate is enforced by the application.

AI disclosure: every rule pack saved by Digesto carries a machine-readable flag indicating it was AI-assisted. Downstream tools can surface this disclosure to end users.

Chunking: large documents are automatically split into chunks to fit within the AI model's context window. Each chunk is processed in sequence; results are merged before review.

Browse Rules

  1. Click Browse Rules. Digesto runs a git pull to fetch the latest packs from your configured repository.
  2. Navigate the domain tree (14 regulatory domains) or use keyword search.
  3. Select a rule to view its full structured definition — ID, domain, standard, section, clause, patterns, status, severity, weight, and remediation.

Use this before ingesting a new document to avoid duplicating rules already captured.

Compliance Graph

Click Compliance Graph to see a GDI+ diagram mapping rule packs to the AvantLeap tools that consume them (e.g. Norma for LEED credit requirements, Collaudo for model-integrity rules). Use it to understand cross-tool coverage and the impact of changing or removing a pack.

Rule & Ecosystem Map

Click Rule & Ecosystem Map to open a live five-column diagram of the full compliance ecosystem:

Column Contents
Doc Sources Document types Digesto ingests (PDF, TXT, MD)
Digesto The Rule Store at the centre of the ecosystem
Compliance Patterns P1–P13 patterns with tier (T1 Native / T2 Delegated / T3 Evidence) and facet
Gatekeepers Tools that query Digesto directly (Custode, Norma, Collaudo, Percorso, Calcolo, Licenza)
Providers Tools that compute data and feed it to gatekeepers (Spazio, Involucro, Registro, Agenda, Sagoma, Conduttura)

Live status (progress %, target date) is fetched from the AvantLeap Docs repository when the dialog opens. If the fetch fails, a built-in offline snapshot is used. The status bar shows which source is active and when the data was last updated.

How Other Tools Consume Digesto

Consuming tools currently read rule packs through the in-process RuleQueryService. A future avantleap-digesto MCP service (port 7324) will serve rule packs over a network API. Either way, the rule packs Digesto produces are the single source of truth — making the compliance suite jurisdiction-agnostic and forward-compatible with new code editions.

Consumer Integration
Norma (22) LEED credit requirements via Digesto query instead of hardcoded values
Collaudo (27) Model-integrity rules per standard
Percorso (38) Distance and path compliance thresholds
Calcolo (37) Performance calculation thresholds
Custode (23) Parameter naming and identity rules
Future compliance tools Any AvantLeap product needing regulatory data

Data Storage

%APPDATA%\AvantLeap\Digesto\
  settings.json            <- config + DPAPI-encrypted secrets (never committed)

<rule-packs git repo>\
  rule-packs\<domain>\*.json   <- structured rules, one folder per domain
  rule-packs\<domain>\*.md     <- human-readable wiki pages

Secrets are DPAPI-encrypted and tied to your Windows user. Rule packs live in git — every change is versioned and attributable.

Tips

  • Always review AI extractions. The draft → validated gate and schema validation are safeguards, not substitutes for human review by a qualified professional.
  • Use Ollama for sensitive documents. Documents that must not leave your network can be ingested with Ollama — no data is sent to any external service.
  • Use least-privilege tokens. Scope your GitHub PAT to only the rule-packs repository.
  • Browse before you ingest. Check Browse Rules first to avoid duplicating rules already captured.
  • Secrets stay on your machine. Copying settings.json to another user or PC yields empty secrets by design — re-enter keys in Settings on each machine.

Troubleshooting

Issue Solution
App won't start Confirm the .NET 8 desktop runtime is installed; download from microsoft.com/dotnet
Git push/pull fails Check the rule-packs repo URL and that your fine-grained PAT is valid and scoped to that repository
AI extraction times out (cloud) Verify the API key in Settings; large documents chunk automatically but each chunk takes time
AI extraction fails (Ollama) Confirm Ollama is running (ollama serve), the base URL is correct, and the model is pulled (ollama pull llama3.1)
Ollama responses are slow Ollama uses local compute — allow up to 5 minutes per chunk on CPU-only machines
Secrets appear empty after copying settings Expected — DPAPI ties secrets to the Windows user; re-enter them in Settings on each machine
Rule validation errors on save Read the error detail — missing required fields or incorrect enum values are the most common causes
Rule & Ecosystem Map shows offline data GitHub fetch failed or timed out — check internet access; the offline snapshot is used as fallback