Research the law.
Check the authority.

How to use the law workspace

1. Sign in through the AI website and open Law research. It uses your existing AI session and password.
2. Ask a focused research question, such as “What sources address California contract consent and consideration?” Local mode searches only the collected legal corpus and returns cited excerpts.
3. Check each original source, the applicable jurisdiction, effective date and whether it is statutory authority or agency guidance. A retrieval timestamp does not establish that law is currently effective.
4. The owner can collect additional configured sources or request a comparison from ready external APIs. Other registered users can search and retain their own private exchanges.
5. Read each AI draft separately. Agreement between providers does not validate a citation or conclusion. A missing search result does not mean no rule applies.

Legal sources are stored separately under AI/law/data. General AI sources are not silently treated as legal authority. Your last 20 completed questions and responses are private to your account; pending consultations remain until processed. Public-source automatic study notes are shared research material. Generated text never enters the authoritative corpus or automatic weight training.

Initial scope and official sources

The configured starter collection covers California Civil Code contract provisions, California business formation guidance, US Small Business Administration guidance and Federal Trade Commission business guidance. It is a small research starting point, not comprehensive legal coverage or a case-law citator. It does not track every amendment, local rule, controlling decision or filing deadline.

• California Courts: breach of contract and debt defenses — court self-help guidance.
• California Civil Code section 1550 — statutory contract formation provision.
• California Civil Code section 1636 — statutory contract interpretation provision.
• California Secretary of State: starting a business and entity types — agency guidance, not a replacement for statutes.
• SBA: launch your business — federal agency business guidance.
• FTC: advertising and marketing on the internet — federal agency guidance.

The crawler respects robots rules, only fetches exact configured HTTPS URLs and refuses redirects, private destinations, oversized responses and exhausted budgets. A source may be skipped. Read collection reports rather than assuming every configured URL was collected. Source HTML may contain navigation and boilerplate; excerpts are not guaranteed to capture the relevant exception.

Consulting other AI services

Connectors are supplied for Gemini, Groq, Hugging Face, OpenAI and a user-owned OpenAI-compatible HTTPS machine endpoint. This is not a promise that every service offers free access to your account. Free websites and paid chat subscriptions are not automatically API entitlements. The server does not scrape chat websites, reuse browser cookies, rotate accounts or evade quotas.

Before enabling a connector, inspect its current account quota, data-use terms and billing settings. Set conservative request limits below your actual allowance. Free status is a manual account declaration: these scripts cannot make a billed account free or inspect every provider's billing controls. Configuration reviews expire after 30 days and must be renewed after checking the account again.

cd /home1/leephdco/DEVEL/AI
# Edit the provider entry, without putting a key into this JSON file:
nano providers.json
# Store a key privately, avoiding shell history and visible echo:
umask 077
read -r -s -p 'Groq API key: ' provider_secret
printf '\n'
printf '%s' "$provider_secret" > private/providers/groq.key
unset provider_secret
chmod 600 private/providers/groq.key

For a free connector, configure enabled: true, its exact model, mode: "free", verified_free: true, billing_disabled: true and today's reviewed_on in YYYY-MM-DD format. These flags record checks you actually performed. Set automatic_public_research: true only if you want scheduled public-source consultations. Each connector permits at most 2 calls/day and 20/month, and produces at most 512 output tokens/request. Failures count and there are no retries. Names of key files are gemini.key, groq.key, huggingface.key, openai.key and machine.key.

On the law page, select the configured free group and explicitly authorize sending the current question and bounded public source excerpts. Ready services are shown before submission. The queue worker saves each draft separately under your user ID. Automatic runs send only a generic research question and public-source excerpts; they never send conversation history, uploaded documents or account data. Do not submit sensitive or privileged facts to an external provider without understanding the disclosure.

Paid access later

Leave allow_paid: false for zero paid calls. To enable deliberate paid comparison later, configure the chosen provider's mode: "paid", actual input/output token prices per million, key and model; set allow_paid: true and a positive monthly_paid_budget_usd no greater than 5. Configure a matching provider-side limit when available. The application reserves an estimated worst-case token cost before each call using your supplied prices; incorrect prices or outside-account activity invalidate the estimate. Paid services are always excluded from automatic learning, even when manual paid comparison is enabled. A provider subscription is not purchased by these scripts.

Connecting your own machine

Run an authenticated OpenAI-compatible model service on hardware you control, with a valid HTTPS certificate and a public hostname reachable from JustHost. Set endpoint to its exact /v1/chat/completions URL and owned_endpoint_confirmed: true. Put its token in machine.key and choose its model. The client rejects HTTP, credentials embedded in URLs, unexpected ports, private network addresses and redirects. It does not expose a local model daemon or open firewall ports. A private machine requires a separately designed secure connection; localhost on JustHost is not your desktop.

Provider references: Gemini pricing, Gemini limits, Groq limits, Groq billing, Hugging Face credits and billing, OpenAI Responses API. Consult each provider's current privacy and data retention terms before activation.

Bounded continuous learning

“Continuous” means repeated short jobs, not a process consuming resources all day. A cron entry starts at minute 17 every six hours in the hosting server's timezone. The general and legal workspaces alternate, so each is normally checked twice daily. The existing five-minute queue worker handles explicit online requests independently, but both use the same lock. A busy lock causes the scheduled invocation to exit rather than run concurrently.

These are conservative application limits, not a guaranteed JustHost entitlement. Other account applications, HTTP concurrency, provider traffic and server overhead also consume resources. Lower the limits if the host reports resource pressure. The scheduler checks its last start time to avoid duplicate early runs; counters and state survive reinstalls. Sources default to a seven-day refresh interval. Unchanged source sets skip retraining. Failed sources remain visible in reports.

cd /home1/leephdco/DEVEL/AI
# Run legal source collection deliberately:
bash launch-ai.sh law-run
# Inspect the most recent scheduled run:
cat data/scheduler.json
# Run a due scheduled cycle (does nothing if not due):
bash launch-ai.sh scheduled
# Pause automatic learning: edit schedule.json and set enabled to false.
nano schedule.json
# Resume by setting enabled to true; existing budgets are preserved.

No API keys are required for source collection, BM25 retrieval, one-shot evaluation or the small statistical model. If no verified free connectors are configured, scheduled study records that no consultations occurred. External AI output is stored in data/study.json or law/data/study.json as unverified notes; it is never mixed into legal sources or fed back into automatic model training. This prevents model-generated assertions from being recycled as evidence. These jobs do not train a foundation model or modify the provider's model weights.

Repeatable setup and private files

cd /home1/leephdco/DEVEL/AI
bash install-ai.sh
# To recreate the isolated Python runtime as well:
bash install-ai.sh --rebuild
bash launch-ai.sh check

The embedded installer contains the law interface, connector code, starter source configuration, tests, guides and scheduler. It preserves existing AI users, law conversations, source data, provider keys and usage counters. Law settings live in law/config.json; API settings in providers.json; schedule settings in schedule.json. No credential is embedded in the Bash file. Private results live in law/private/requests.sqlite; API reservations live in private/provider-budget.sqlite. Save logs and checks are available in the private root logs directory and behind login.

If a page fails, inspect installation reports and run the checks. If a consultation fails, verify the provider's key, exact model, free/billing status, review date, limits and destination. Never disable TLS verification or clear usage counters to work around a quota. In a legal emergency or approaching deadline, contact a qualified lawyer rather than relying on this scheduled research tool.