VALORDB

Veterans Appeals Library & Outcomes Research — in your terminal.

A sovereign, grounded CLI and Python client over every published Board of Veterans' Appeals (BVA) decision (1992–present, ~1.86 million) wired to the Title 38 statutes and regulations (38 U.S.C. + 38 CFR) that govern them.

pip install valordb
valordb

Why this is different: the model is never the source of a fact

Legal research tools that let an LLM free-text statistics and citations are a liability — the model will confidently invent a grant rate or a regulation that doesn't exist. VALORDB makes that structurally impossible:

• Statistics are computed by the corpus, injected into the prompt as a locked VERIFIED_FACTS block, and any number the model emits that isn't traceable to that block is struck after generation.
• Citations are verified against the actual Title 38 store — anything that doesn't resolve is flagged, and M21 manual cites outside the mirror are flagged (VALORDB mirrors only part of M21).
• The model's job is reduced to explaining fixed facts. It may not compute, infer, rank, or invent a statistic or a citation.

The grounded analyst runs on your own Ollama — nothing leaves your machine.

Quick start

Interactive (a research workstation)

valordb

  ██╗   ██╗ █████╗ ██╗      ██████╗ ██████╗ ██████╗ ██████╗
  ██║   ██║██╔══██╗██║     ██╔═══██╗██╔══██╗██╔══██╗██╔══██╗
  ╚██╗ ██╔╝██╔══██║██║     ██║   ██║██╔══██╗██║  ██║██╔══██╗
   ╚════╝  ╚═╝  ╚═╝╚══════╝ ╚═════╝ ╚═╝  ╚═╝╚═════╝ ╚═════╝
  VALORDB CLI — valordb.org

valordb> ask PTSD and sleep apnea claimed together, under Title 38
valordb> conditions sleep apnea
valordb> law usc-38-1110
valordb> upgrade

One-shot / scriptable

valordb ask "overview of Sleep Apnea + PTSD claims" --mode attorney
valordb kpi --json | jq .rates
valordb conditions "sleep apnea"
valordb dimension judge
valordb law cfr-38-3.310

As a Python library

from valordb import VALORDB

db = VALORDB(api_key="vk_firm_…")          # or set VALORDB_API_KEY
print(db.kpi()["rates"])                    # corpus headline rates
c = db.conditions()
print(c.rate("sleep apnea"))                # 0.4171  (any-grant)
print(c.together("PTSD", "sleep apnea"))    # real co-occurrence rate + N
print(db.law("usc-38-1110")["heading"])     # a Title 38 section (free)

Commands

Tiers

The Title 38 & M21 law is free to read (here and on valordb.org). The analytics, decision search, grounded AI, and API/CLI are paid:

valordb upgrade runs a Paddle checkout from the terminal and saves your key. Veterans and non-profits — discounts available; email joseph@resilientmindai.com.

Configuration

Precedence: environment variable → ~/.valordb/config.toml.

export VALORDB_API_KEY=vk_firm_…
export VALORDB_OLLAMA=http://192.168.0.92:11434   # a LAN Ollama box works from the CLI

The key is stored at ~/.valordb/config.toml (chmod 600) and never logged.

For agents & automation

The CLI is a thin client over the keyed REST API (/api/v1, Firm+). Anything that can send an X-API-Key header — including an Ollama tool/agent or an MCP server — can call the same endpoints directly (JSON in, JSON out). Use --json for machine output. Endpoints: status, kpi, conditions, kpi/dimension/{dim}, law/{id}.

Sovereignty & privacy

• The grounded ask runs entirely on your Ollama (local or LAN). Your question and the corpus context never leave your machine.
• The optional httpx extra (pip install valordb[httpx]) uses httpx instead of urllib for API calls. The importable SDK only needs requests.

Legal

Legal information, not legal advice. VALORDB surfaces public decisions, statutes, and regulations and how the Board has historically ruled. It does not interpret your claim and is not a substitute for an accredited representative or attorney. VA claims carry hard deadlines and one-shot effective-date consequences — verify everything against the controlling law and accredited help (va.gov/ogc).

The BVA corpus is public record. Per-judge and per-office rates reflect docket and case mix, not a controlled measure of adjudicator favorability.

Created by Joseph C. McGinty Jr. · © 2026 ResilientMind AI LLC · valordb.org