AI legal research that runs on your servers, not someone else's.
CaseCite is a free, source-available legal AI platform your firm hosts itself. Ask questions of your own documents and get answers cited back to the source, review and redline contracts, draft long agreements from your precedents, and research cases and judges. Your documents, search index and chat history stay on infrastructure you control, and the AI runs on providers you choose, under your firm's own keys.
Most legal AI tools start the same way: upload your clients' contracts, pleadings and correspondence to the vendor's cloud, accept the vendor's retention policy, and pay a per-seat price for as long as you want access to your own work product.
CaseCite turns that around. The software runs on a server your firm controls, and your documents, the search index, chat history and every analysis are stored there. The AI providers are the ones you pick, called with keys your firm holds. Nothing is ever sent to the CaseCite maintainers, and there is no telemetry. If you want document text never to leave your network, point CaseCite at models you host yourself.
Your documents, your servers, your keys. The AI works for the firm, not the other way around.
What CaseCite does for your firm.
What stays on your servers, and what leaves.
Everything your firm creates
CaseCite keeps its data on the deployment you run. Nothing is stored with the CaseCite maintainers, because nothing is sent to them.
- Uploaded documents, with the originals encrypted at rest
- The search index built from them
- Chat history, contract analyses, redlines and drafts
- The audit log, user accounts and encrypted API keys
Just what each task needs
The AI features send content to providers your firm picks, under your firm's own accounts. Each flow is named, so you can explain it to a client.
- Uploading a document sends its text to your embeddings provider: OpenAI by default, or Voyage or Cohere
- Asking, analyzing or drafting sends the question and the relevant passages to your chat model: OpenAI, Anthropic or Gemini
- The case-law tools send search queries and citations to CourtListener
- To the CaseCite maintainers: nothing. No telemetry, no phoning home
Want document text to stay entirely on your network? Run both models yourself: point chat (OPENAI_BASE_URL) and embeddings (EMBEDDING_BASE_URL) at OpenAI-compatible servers you host. An allowlist of approved AI providers is enforced on every call, and the search index stays local unless you choose Pinecone. Every outbound data flow is listed in docs/subprocessors.md.
A 40-page draft lands at 9:00. Your markup is ready before lunch.
- 09:00 Import Opposing counsel's draft services agreement comes in, and you pull it from your document system into CaseCite.It's indexed alongside your firm's own forms and past deals. Indexed
- 09:04 Review CaseCite reviews it against your practice profile and flags the risky clauses and missing protections, ranked by severity. Flagged
- 09:15 Compare You compare it clause by clause against your firm's standard agreement and see exactly where it departs. Compared
- 09:40 Redline You export the redline as Word tracked changes, ready to send back.A fidelity check refuses to produce a redline whose text doesn't match the original document. Redlined
- 10:30 Verify You check each change against the source before it goes out.The analysis carries a verify notice, and the work is recorded in the signed audit log. Signed off
AI is fast, and sometimes wrong. A lawyer can't afford to be. So CaseCite is built to make checking quick, and to say so plainly when it can't be sure. It's a research aid, not a substitute for your judgment or your citator.
- Cited to the source
- Research answers cite the passages they came from, so you can check every statement against the document behind it.
- Honest citation check
- It scans citing opinions for negative treatment. If it finds any, it won't call the case good law, and it says plainly that it isn't a Shepard's or KeyCite substitute.
- Verify notice
- Every answer, analysis, redline and draft carries a notice to verify it against the sources and current law, on screen and in every Word, PDF and Markdown export.
- Hard to hijack
- Documents, opinions and contracts reach the AI as untrusted content. Instructions hidden inside them are ignored and pointed out, never followed.
- Faithful redlines
- A character-level check refuses to produce a redline whose text differs from your original document.
- Nothing skipped quietly
- Contract summaries and authority maps read the whole document and tell you if any part was left unread.
Security your managing partner can sign off on.
Originals and keys locked with AES-256-GCM
Uploaded documents, API keys, connector tokens and two-factor secrets are encrypted by the application. Everything else sits on a disk you encrypt, and CaseCite won't start until you confirm it is.
A tamper-evident record of who did what
Every entry is HMAC-signed and chained to the one before it, then verified at startup and on demand. Admins can filter it, check its integrity and export it.
Sign-ins that expire on their own
Authenticator-app two-factor with recovery codes, mandatory if you choose. Sessions end after 8 hours or 2 hours idle, and "sign out everywhere" ends every session at once.
Everyone sees only their own work
Role-based permissions an admin can adjust, with every document, chat and analysis scoped to its user and matter. Self sign-up is off; colleagues join by invite.
Your AI keys never reach the browser
Provider keys are encrypted on the server and only ever shown masked. Each attorney can bring their own key for per-user cost tracking, or the firm can set one shared key.
Only the AI vendors you approve
An allowlist of approved AI providers is checked on every chat and embedding call, so client text can't be routed to a service the firm hasn't vetted.
An unsafe install won't boot
CaseCite refuses to start with missing or placeholder secrets, and the first admin account can only be claimed with a setup token from your server's configuration.
Encrypted backups you can restore
Backups are encrypted by default, checksummed, and verified in full before a restore touches anything.
Running on your own machine in three commands.
$ git clone https://github.com/thomasunise/casecite.git && cd casecite
$ python3 scripts/setup_production.py --create-env --domain localhost --disk-encrypted
$ docker compose -f docker-compose.prod.yml up -dUse --disk-encrypted only if your disk is encrypted (FileVault, BitLocker or LUKS). Open https://localhost, accept the first-visit certificate warning, and register the admin account with the setup token REGISTRATION_BOOTSTRAP_TOKEN from .env.
Then save an AI provider key under Settings, and a free CourtListener API token to unlock the case-law tools. Going to production? Point a domain at your server and the bundled proxy fetches the certificate for you. Keep a copy of SECRET_KEY, ENCRYPTION_SALT and AUDIT_HMAC_KEY off the server, because backups can't be restored without them. The full guide is in docs/deployment.md.
Do it yourself, or have it running this month.
Free for your firm
Clone it, deploy it, adapt it. Best for firms with IT staff who want to run every piece themselves.
- Free to self-host and modify for your firm's internal use under the Elastic License 2.0
- Deployment, secret-rotation, incident-response and data-flow guides in the repo
- Bug reports and feature requests welcome on GitHub
I install it and hand it over running
For firms that want private AI without becoming an IT shop. You get a working system your attorneys can use, not a project.
- Deployed on your own server or private cloud. You own the server, the data and the keys
- Connected to your document systems, with your forms, precedents and past deals indexed
- AI set up under the firm's accounts, or on models you host so document text never leaves your network
- Two-factor, roles, the audit log and encrypted backups configured and tested
- Practice profiles drafted with your attorneys, so reviews and drafts start from how you work
- Attorneys and staff trained on research, review and drafting, and on verifying what the AI produces
- An optional support contract after go-live
CaseCite questions, answered.
What is CaseCite?
CaseCite is a free, source-available legal AI platform that your firm hosts itself. It puts eight tools behind one login, all reading from your firm's documents: research chat with citations, document management with imports from your existing systems, contract review and redlining, long-document drafting, a case view for chatting with an opinion, an authority map, judge analytics, and case-law tools backed by CourtListener.
Does our clients' data leave the firm?
Documents, the search index, chat history and analyses are stored on your own server. Using the AI features sends content to the providers you choose: a document's text goes to your embeddings provider when you upload it, and your question plus the relevant passages go to your chat model when you ask, analyze or draft. The case-law tools send search queries and citations to CourtListener. Nothing is sent to the CaseCite maintainers, and there is no telemetry.
Can we keep everything on our own network?
Yes. Run both models yourself by pointing chat and embeddings at OpenAI-compatible servers you host, and no document text leaves your network. Set both: pointing only the chat model at your own server still sends document text out for embedding. The case-law tools still query CourtListener, a public database, with the searches and citations you look up.
Which AI models does it work with?
OpenAI, Anthropic Claude or Google Gemini for chat, and OpenAI, Voyage, Cohere or a server you run for embeddings. Any OpenAI-compatible endpoint works too, including a model you host yourself. An allowlist lets the firm approve exactly which providers can be called.
Does it replace Westlaw or Lexis?
No, and it doesn't pretend to. Its case-law source is CourtListener, the free public database, and its citation check scans citing opinions for negative treatment rather than acting as a full citator. Where CaseCite earns its place is your firm's own work product: the briefs, contracts, forms and past matters you've built up. Most firms will run it alongside their existing research subscription.
Can the AI make things up?
Any AI model can be wrong, so CaseCite is built to make checking fast. Research answers cite the passages they came from in your documents. Every answer, analysis, redline and draft carries a notice to verify it against the sources and current law, including in exports. Documents and opinions are treated as untrusted, so instructions planted inside them are ignored. It's a research aid: an attorney still verifies the work before anyone relies on it or files it.
What does "source-available" mean? Is it open source?
CaseCite is licensed under the Elastic License 2.0. Your firm can self-host it for internal use and modify it for your own deployment, free. The license doesn't allow offering CaseCite to others as a hosted or managed service, or removing its license notices. Because of those limits it's called source-available rather than open source in the strict OSI sense, but you can read every line of the code.
Which document systems does it connect to?
Google Drive, OneDrive, Box, Dropbox, iManage, NetDocuments, Filevine and Clio, through a sign-in sync or a file picker, plus direct upload. Microsoft and Google connections ask for read-only access. The iManage and NetDocuments connectors are built to the vendors' published APIs but haven't yet been validated against a live account, so test them in your environment first. That testing is part of every install I do.
Is it secure enough for privileged material?
It's built with the controls a firm's IT review asks for: AES-256-GCM encryption of original documents and keys, two-factor sign-in, role-based access scoped to each user and matter, session limits, a tamper-evident audit log and encrypted backups. Two things to know. Derived text, such as the search index and chat history, relies on disk encryption that you provide, and CaseCite won't start until you confirm it's in place. And CaseCite holds no security certification such as SOC 2, ISO 27001 or HIPAA. The repo's SECURITY.md states plainly what it does and doesn't do.
Is it ready for production?
The core product (research, documents, contracts, drafting and the case-law tools) has a full interface and is tested in CI, including a production-mode smoke test. Some pieces are earlier: matters and strategy briefs work through the API but don't have a screen yet, single sign-on is limited to Azure AD on the backend, and ethical walls and legal holds are on the roadmap rather than in the product. The repo's ROADMAP.md lists every known gap. If one of them matters to your firm, ask on the call and you'll get a straight answer.
What does it cost?
Self-hosting for your firm's internal use is free. You pay only for the server you run it on and the AI usage on your own provider accounts, with no per-seat fee. If you want it installed for you, the install is scoped and priced after a short call, and you keep everything: the code, the configuration and the keys. Support contracts, and licenses to host CaseCite for other firms, are available separately.
We're a small firm without an IT department. Is this for us?
If your firm works from documents and cares where client files go, it's built for you. You don't need to run it yourself. Book a call, tell me how your firm works and what you use today, and I'll tell you plainly whether CaseCite is the right fit or whether something simpler would serve you better.
Private legal AI on your own servers, running this month.
Thirty minutes. We look at how your firm works, where your documents live and what an install takes. You leave with a plan, a timeline and a fixed price, or a straight answer that you don't need it yet.
