Frontier Control
Use the best AI. Keep control of the firm.
Your lawyers already work in ChatGPT, Claude, Grok, and the frontier assistants that will come next. Now they can work with EdgeLex from inside them.
Ask about a matter. Find the controlling document. Review email. Trace a deadline. Draft a response. Create a task. Prepare a calendar event. Verify a citation. Propose the next step.
The assistant is where your lawyers work.
EdgeLex is where the files, the rules and the record stay.
Your AI. Your interface. Your firm’s rules.
What the firm keeps
Change the AI whenever you like. Your firm’s rules stay put.
Frontier models move fast. You should be able to use the best one available without rebuilding how your firm works around each provider.
EdgeLex connects your firm’s workspace to ChatGPT, Claude, Grok, and other MCP-compatible AI clients. Each lawyer picks where to work. The firm still controls:
- Identity and permissions
- Matter access
- Documents and evidence
- Approved tools
- Action boundaries
- Human approvals
- Audit history
- The system of record
The AI provider brings the interface and the intelligence. Your firm keeps the keys.
The assistant asks EdgeLex for what it needs. EdgeLex decides what it gets.
Most integrations copy your information into an AI product and hope the prompt keeps it under control. EdgeLex works the other way round.
The assistant never becomes your document system, matter database, task list, calendar, billing record or workflow engine. It asks EdgeLex for what it needs, and EdgeLex checks who is asking, which matter, which record, and what they want to do with it.
You get back only what you are allowed to see. Only approved actions go through. Every result stays in EdgeLex, where the firm can see it.
The model can reason. EdgeLex decides.
Ask from the interface you already use.
From a supported frontier assistant, a lawyer can ask:
“Show my active matters.”
“Read the filed brief in Wiggins v. Duncan and prepare me for mediation.”
“Find every email in this matter discussing the missing discovery responses.”
“What deadlines are coming up, and which court rules control them?”
“Draft a response to opposing counsel and attach the latest reproduction.”
“Create a task to file the opposition next Friday.”
“Build a standing agent that briefs me whenever a new filing enters this matter.”
The assistant can read and reason across the EdgeLex capabilities the firm has authorized. It does not receive universal access merely because it is connected.
Lex can read everything you can. It cannot change anything without you.
EdgeLex keeps four steps separate. Most AI products blur them together:
Read
The assistant can see only the matters, documents, email, calendar entries, tasks, billing records and court rules you are allowed to see.
Analyze
It can search, compare, summarize, extract, calculate and reason over what it is allowed to see, and it keeps the sources attached.
Propose
When it spots a useful next step, it writes up a proposal: send this message, create this task, update this event, save this document. Nothing has happened yet.
Execute
You see the exact action and approve or decline it. EdgeLex checks your permissions and the current state again before it runs, then records what it actually did.
Nothing happens until you say yes. When you do, EdgeLex checks again and records exactly what it did.
That is the difference between an AI demo and a system a law firm can run its practice on.
The approval lives where the lawyer is working.
You should not have to leave the conversation, open another application, and work out what the AI meant to do. EdgeLex shows the proposed action right in the chat:
You confirm it in plain sight. EdgeLex carries it out on the server. The audit trail stays with the firm.
When something is missing, EdgeLex stops and asks.
- Possessing a matter ID is not permission to read it.
- Mentioning a colleague's first name is not enough to send a message.
- A model's conclusion is not a court deadline.
- A suggested task is not an approved task.
- An old object version cannot silently overwrite a newer one.
- A standing agent cannot activate itself.
EdgeLex checks who is asking, enforces matter access, checks the evidence, catches stale edits, applies workflow rules, and stops when a fact or a permission is missing.
You get the full power of the model. The firm keeps the keys.
177 governed tools
One connection. The legal operating system behind it.
Everything Lex can do in EdgeLex, it can do from the assistant you connect:
Matters
Context, timelines, active matters, legal instruments, events, and status.
Documents
Search, text retrieval, fact extraction, version comparison, court compliance, forms, and governed saving.
Search, threads, attachments, drafts, matter linking, organization, and authorized sending.
Calendar and deadlines
Events, availability, scheduling, invitations, court-rule calculations, and deadline lineage.
Tasks and workflows
Assignments, completion, case roadmaps, blockers, approvals, and governed transitions.
Research and citations
Case-law search, opinion retrieval, citation verification, proposition support, and cite-check reports.
Billing
Time, expenses, invoices, matter arrangements, and reviewable billing proposals.
Messages, meetings, depositions, notes, clauses, and firm intelligence
One permission model. One matter context. One audit trail.
Native to EdgeLex and part of every subscription: lawyers use their own frontier accounts; nothing extra to buy. For IT and security reviewers, the full model is on the technical capability and governance page: authorization layers, read vs direct vs proposal-first writes, evidence-bearing writes, stale-write protection, workflow and court-rule controls, standing agents.
One connection, one permission model, one audit trail.
One legal operating system, reached through whichever assistant your lawyers prefer.
Give Lex a standing job, and keep your hand on the switch.
Describe a recurring job in plain language:
“Whenever a new document is added to this matter, send me a five-bullet brief and identify every date it mentions.”
EdgeLex turns the request into a design your firm has already authorized. You review it. Creating it needs your confirmation. The new agent starts switched off, and turning it on is a separate decision.
Every run stays inside the firm’s permissions, tools, evidence rules, spending limits, approval gates and audit trail.
Agents can work around the clock. Their authority never grows on its own.
Switch models whenever a better one ships.
Use ChatGPT today. Claude for a different workflow. Grok where your policy permits it. A local model for sensitive work. Your own firm model for institutional knowledge.
When the interface changes, your matters do not move. Your document history stays whole. Your task list stays in one place. The approval step stays. EdgeLex is the layer underneath that does not change.
Your matters, documents and history stay exactly where they are.
What the assistant sees, and who decides.
When a lawyer asks an external frontier AI to work with EdgeLex, the information EdgeLex returns may be processed by that provider under the firm’s account, subscription, configuration, and provider terms. EdgeLex does not pretend otherwise.
Instead, the firm controls which users may connect, which tools are available, which matters and records are accessible, and which actions require approval. Firms can choose an enterprise frontier account, restrict available capabilities, use private deployment, or keep sensitive work on approved local models.
We would rather tell you where the line is than pretend there is none.
Questions
- Which assistants work with EdgeLex?
- Any MCP-compatible client: ChatGPT, Claude, Grok, and the frontier assistants that come next. The lawyer signs in once; the assistant then works with the 177 governed EdgeLex tools the firm has authorized for that user.
- Does connecting an assistant give it access to everything?
- No. The assistant can retrieve only the matters, documents, email, calendar records, tasks, billing records and court rules the authenticated lawyer is permitted to access. Possessing a matter ID is not permission to read it.
- Can the assistant take actions on its own?
- No. It can write up a proposal: send this message, create this task, update this event, save this document, advance this workflow. Nothing happens until you confirm the exact action. EdgeLex checks your permissions and the current state again before it runs, then records what it actually did.
- Where does the information go?
- Information EdgeLex returns to an external assistant may be processed by that provider under the firm's account and terms. The firm controls which users may connect, which tools are available, which matters are accessible, and which actions require approval; firms can use an enterprise frontier account, restrict capabilities, use private deployment, or keep sensitive work on approved local models.
- What does it cost?
- Frontier Control is native to EdgeLex and part of every plan, not a third-party connector. Usage on the assistant side is billed by that provider under the firm's own account.
- Is there a technical description for our IT team?
- Yes: the technical capability and governance model at edgelex.com/platform/frontier-control/technical covers the 177-tool inventory by domain, the authorization layers every call passes through, read-only versus direct versus proposal-first writes, evidence-bearing writes, stale-write protection, workflow and court-rule controls, standing-agent lifecycle, and the rules an AI must follow.
See also Lex AI & Agents, AI Governance and Deployment.
Bring the AI your lawyers want into the system your firm controls.
Your lawyers work in the assistant they like. The files, the rules and the record stay in EdgeLex.