A sovereign, grounded CLI + Python client over every published Board of Veterans' Appeals decision (1992–present, ~1.86M), wired to Title 38 (38 U.S.C. + 38 CFR). Ask corpus-locked questions, pull outcome rates, and read the law — without leaving the shell. The grounded analyst runs on your Ollama; nothing leaves your machine.
$ pip install valordb && valordb
Why it's different
Most "AI for veterans" tools let a language model invent numbers and citations. VALORDB inverts that: statistics are computed by the corpus and locked, citations are checked against the real Title 38 store, and inference is yours to run.
Every rate is computed by querying the actual corpus and injected as a locked
VERIFIED_FACTS block. Any number the model emits that isn't traceable to
that block is struck after generation — it cannot pass off a guess as data.
Every 38 U.S.C. / 38 CFR citation the model emits is
resolved against the ingested Title 38 store. Citations that don't resolve are
flagged — so you never chase a phantom regulation.
The grounded ask runs entirely on your own Ollama
(localhost or LAN). Your question and the corpus context never leave your
machine — no third-party AI provider sees a thing.
In practice
Four moves cover most of it: install, ask a grounded question, pull the headline rates,
and read the controlling law. Each block below is a representative valordb
session — the exact commands you'll run and the shape of the output (figures shown are
illustrative).
$ pip install valordb Successfully installed valordb-0.5.0 $ valordb __ ____ _ ___ ____ ____ ____ / / / / _ |/ / / _ \/ __ \/ __ \/ __ ) \ \/ / __ / /__/ // / /_/ / /_/ / /_/ / \__/_/ |_\___/____/_____/_____/_____/ VALORDB CLI — valordb.org Type a command, or `help`. `quit` to exit. valordb> _
First run: the install line, then the VALORDB CLI banner.
valordb> ask PTSD and sleep apnea claimed together, under Title 38 # grounding on local Ollama · corpus-locked stats injected… When obstructive sleep apnea is claimed secondary to service-connected PTSD, the Board's any-grant rate for the sleep-apnea issue is 37.1% (N=12,480), versus 29.3% corpus-wide. Secondary service connection turns on a medical nexus under 38 CFR 3.310 ✓ verified. Sources: VERIFIED_FACTS (corpus) · 38 CFR 3.310 · 38 U.S.C. 1110 valordb> _
A grounded answer — every figure traced to the corpus, every citation verified.
$ valordb conditions "sleep apnea" Sleep apnea · any-grant 31.8% (N=88,204) AMA lane any-grant 34.2% (N=21,019) Legacy any-grant 31.0% (N=67,185) Top co-occurrence: PTSD · hypertension · diabetes mellitus II $ valordb kpi --json | jq .rates { "any_grant": 0.293, "grant": 0.117, "remand": 0.234, "deny": 0.247 } $ _
Per-condition rates with N, plus scriptable JSON for pipelines.
$ valordb law cfr-38-3.310 38 CFR § 3.310 — Disabilities that are proximately due to, or aggravated by, service-connected disease or injury. (a) General. …a disability which is proximately due to or the result of a service-connected disease or injury shall be service connected.… # reading the law is FREE — no key required $ _
Verbatim Title 38 — reading the law is free, no key required.
Quick start
Run it as an interactive REPL, fire one-shot scriptable commands, or import the Python client. Install once: pip install valordb.
Run valordb with no arguments. You get an ASCII-art banner and a valordb> prompt.
$ valordb # then, at the valordb> prompt: valordb> ask PTSD and sleep apnea claimed together, under Title 38 valordb> conditions sleep apnea valordb> law usc-38-1110 valordb> upgrade
Every command works as a single shell invocation — pipe --json output straight into jq.
$ 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 $ valordb search tinnitus
Import the client for notebooks, pipelines, and apps. Optional HTTP extra: pip install valordb[httpx].
from valordb import VALORDB db = VALORDB(api_key="vk_firm_…") db.kpi()["rates"] db.conditions().rate("sleep apnea") db.conditions().together("PTSD", "sleep apnea") db.law("usc-38-1110")["heading"]
Reference
Reading Title 38, M21, and CAVC is free. Grounded AI, analytics, search, and the API/CLI are paid (Firm+).
| Command | What it does |
|---|---|
ask <q> [--mode veteran|attorney] | Grounded analyst on your local Ollama — corpus-locked stats, verified citations. |
kpi | Headline outcome rates (any-grant / grant / remand / deny). --json for scripting. |
conditions [name] | Per-condition any-grant + N, AMA-vs-Legacy split, co-occurrence. |
dimension <dim> | Breakdown by judge, regional_office, condition, rep_type, … |
law <id> | A Title 38 section, verbatim. FREE (e.g. usc-38-1110, cfr-38-3.310) |
m21 <id> | An M21 manual section. FREE (sub-regulatory; partial mirror) |
search <q> | Search Title 38 + M21 + CAVC by keyword / citation. FREE (full-text + layman synonyms; decision-corpus search is Firm+) |
setup | Guided local-AI install — offers to install Ollama + pull a tiny model. |
config | Configure the AI engine — Ollama (localhost/LAN), BYOK cloud, or Claude Code CLI. |
info | Your key: tier + quota. |
key / login / logout | Show / set / paste / remove your license key. |
upgrade | Subscribe / upgrade — purchase through the CLI (Firm+ for the API). FIRM+ |
site / help / quit | Open the site · reference · exit. |
Access
The Title 38 + M21 + CAVC law is free to read — no key. Analytics, search, grounded AI, and the API/CLI are paid. valordb upgrade runs a Paddle checkout right in the terminal and saves your key.
| Tier | Price | What you get |
|---|---|---|
| Veteran | Free | Read the complete Title 38 + M21 + CAVC (no key). |
| Professional | $49/mo | Web analytics dashboard + Attorney-grade AI. |
| Firm | $149/mo | REST API + this CLI + bulk export + team (5 seats). |
| Enterprise | $499/mo | Full API + white-label + SLA + sovereign on-site appliance. |
Veterans & non-profits — contact us about custom arrangements: joseph@resilientmindai.com. See the full pricing page.
Configuration
Precedence: an environment variable wins, otherwise ~/.valordb/config.toml. Your license key is stored at that path (chmod 600) and is never logged.
| Setting | Env var | config.toml | Default |
|---|---|---|---|
| License key | VALORDB_API_KEY | [auth] api_key | — |
| API base | VALORDB_URL | — | https://valordb.org |
| Ollama host | VALORDB_OLLAMA / OLLAMA_HOST | [ai] ollama_host | http://localhost:11434 |
| Ollama model | VALORDB_MODEL | [ai] model | first installed |
# ~/.valordb/config.toml [auth] api_key = "vk_firm_…" [ai] ollama_host = "http://localhost:11434" model = "llama3.2:3b"
Sovereignty & privacy
The grounded ask runs entirely on your Ollama (localhost or LAN). Your question and the corpus context never leave your machine — no third-party AI provider is involved. Run valordb setup to install Ollama and pull a tiny model in one guided step.
Point the CLI at your own Ollama with VALORDB_OLLAMA or
[ai] ollama_host. Cloud/BYOK is an opt-in convenience — local is the default,
and the privileged path.
Your license key lives in ~/.valordb/config.toml at chmod 600 and is
never written to logs. Reading the law needs no key at all.
Install the faster transport when you want it:
pip install valordb[httpx]. Everything else works with the standard library.