Panorama of Tools (Five Questions That Place Any Tool)

There are hundreds of AI tools for legal work, and each month brings new ones. A list by name tells you what exists. It does not tell you what to choose.

The map above places the best-known tools on five questions. Pick two questions to compare them. The full list of about a hundred tools, with a search, comes after the families below. The five questions are the real content of this lesson. They place any tool, including the tools that are not on the map.

Two warnings before you use the map.

• A position is not a grade. Most of these tools run on the same few models from OpenAI, Anthropic and Google (lesson 4.1). The map shows what a vendor builds around the model: sources, interface, price, control. It does not show which tool gives better answers on your work. Lesson 4.6 is about that test.
• Positions move. In 2026, a workbench bought a case-law database and a publisher agreed to buy an AI start-up (lesson 4.3). The map is a snapshot of October 2026.

The market in eight families

The market falls into eight families. In short:

• General AI. ChatGPT, Claude, Gemini, Copilot, Le Chat. They do any task with text. Many lawyers start here (lesson 1.4).
• Build your own. Coding agents, automation tools and code libraries. You make a tool with them. Alone, they do nothing legal.
• Legal research and platforms. Publishers with their own databases (Lexis+, Westlaw, Doctrine), legal workbenches (Harvey, Legora), and the free official sources (Légifrance, EUR-Lex).
• Contracts and documents. Drafting in Word (Spellbook), review in volume (Kira, Luminance), contract management (Ironclad), signature (Docusign).
• Litigation and disputes. eDiscovery (Relativity, Everlaw), litigation analytics (Lex Machina), and tools for briefs and hearings.
• Running the firm. Practice management (Clio), document management (iManage), time and billing.
• Compliance. Privacy (OneTrust), regulatory change, sanctions screening, whistleblowing.
• Practice-area specialists. One task in one practice: injury demand letters (EvenUp), patent drafting, trademarks, immigration.

Search the list below by name, by task or by country. It shows the best-known tools first. The less-known tools are one tap away.

AI moves to where your documents already are

Look at the families “Running the firm” and “Litigation and disputes”. Practice management, document management, e-signature and eDiscovery are not AI products. They are systems that hold your work. Your matters, your documents and your evidence are already inside them.

Each of these systems now adds AI over what it holds. iManage and NetDocuments add assistants over the firm's documents. Relativity reviews evidence with its aiR tools. Clio went further: it bought vLex, a legal research library, for $1 billion, and joined the library to its matter data Clio · vLex acquisition, Nov 2025 (opens in a new tab).

The reason is simple. A model is easy to rent. Access to your documents, with the correct permissions for each matter, is hard to get, and the system that holds the documents already has it. Before you buy a new AI tool, find out which AI your current systems already include. It is often the shortest path to a first use that your IT team can approve.

Five questions that place any tool

Each axis of the map is one question. Each answer is a band, not a score. A tool in the band “Quote after a demo” can cost less than a tool with a published price. The bands describe how a tool is made and sold. They do not rank tools.

Specialisation buys three other things:

• A fitted workflow. The steps, the forms and the output format of one task, already built.
• Data. Sometimes a database that a general tool cannot reach.
• People who know the task. The vendor has seen the task thousands of times.

A specialist earns its price when your task is narrow and you do it often. An injury firm that sends 300 demand letters a year gets the fit back many times. A firm that drafts one patent a year probably does not. Small specialists are also the vendors most likely to be bought or to close. Ask what happens to your data in that case.

In the map, put “Specialised for law” across and “Legal sources” up, then look at the bottom right. Many tools built for legal work have no legal sources. This is not a defect. A contract-review tool must read your contract, not case law. But do not ask it whether a clause is enforceable, and do not trust a citation that it writes.

The difference between borrowed and owned data also matters. An owner controls the corpus, its updates and its citator. A borrower depends on a licence that can change. That is why, in 2026, Legora bought a case-law database and Harvey added millions of court opinions as its own source.

For any tool, ask the question from lesson 4.3: “Which database does this answer come from, and can I open the source?” Then check the citation in the official source. Légifrance, EUR-Lex and CourtListener are free.

The last rung explains a word that you will hear: headless. A headless tool has no screen of its own. Other software uses it: an API, an MCP server, a code library. It gives a capability, not a place to work. A platform is the opposite: a place where you log in and do the work.

The rungs exchange effort for control. At the first rung, the vendor does the work, and you accept its choices. At the last rung, you get exactly what you need, and you own the result. You also own its maintenance. A tool that one associate built in a weekend stops when that associate leaves (lesson 5.5).

Build only what you repeat and what makes your work different from the work of other firms. Buy the rest. For most lawyers, the effort goes to the second rung. A skill is a text file that you write once. It is not code.

Three points follow.

• The process is a cost. With a published price, you can test the tool this afternoon. An enterprise contract takes weeks or months of procurement. For a solo lawyer or a small firm, the right half of this axis is closed in practice.
• A low price can hide weak terms. On some consumer plans, the provider can use your content to train its models. The plan, not the product name, decides the terms (lesson 3.2).
• The invoice is not the full cost. Add the time to set the tool up, to train the team, and to check each output (lesson 1.5).

The first price to ask for is the price of a test on your own work. A vendor that refuses a test on your own non-confidential documents asks you to buy on trust.

Between the two edges, a few products offer a private deployment: the vendor installs its software on your servers or in your own cloud account. Mistral sells Le Chat Enterprise this way Mistral AI · Le Chat Enterprise, May 2025 (opens in a new tab). iManage can run on a firm's own servers. You get more control over the data. You also take on more of the work to run it.

For a ready-made product, “is it open?” is the wrong question, because the answer is almost always no. Ask instead what you can take with you when you leave. The export questions of lesson 4.3 apply: your skills, your logs and your documents, in plain formats.

How to use the map

Four steps. Each one takes a few minutes.

1. Start from the need. Pick a need in the full list, or type a task: “trademark”, “due diligence”, “billing”.
2. Check what you already pay for. Your research service, your document system and your practice-management software may already include AI. Ask each vendor what your current contract covers.
3. Put your hardest limit across. If you need citations, choose “Legal sources”. If you have no IT team, choose “Built by you”. If you work alone, choose “Price”. If continuity or the location of your data is the issue, choose “Openness”. The tools on the wrong side of that axis drop out.
4. Test two or three tools on your own work. Use the same documents and the same questions for each. Lesson 4.6 gives the criteria, the method of the test and the questions to send each vendor.

An example

Two lawyers in Lyon draft and review commercial leases. They have no IT team and a limited budget. For some points of law, they need French sources.

• Need: “Contracts and documents” for the leases, “Legal research and platforms” for the points of law.
• Hardest limit: money and effort. They put “Price” across and “Built by you” up.
• Result: the enterprise products drop out. Three tools remain: a general assistant on a business plan, with a skill that holds their lease checklist; a French research service with a published price; and Légifrance to check each citation. A contract-review tool stays on the list only if their volume grows enough to justify a quote.

This is not a recommendation for every small firm. It shows the method: the need selects a family, and one limit removes most of that family.

What the map does not show

The five questions describe the shape of a tool. Four important things are not on the map.

• Quality on your work. Only a test with your documents shows it (lesson 4.6).
• Confidentiality terms. Where your data goes, who can see it, how long it stays (lessons 3.2 and 3.3).
• Fit with your tools. Whether it works inside Word, Outlook and your document system.
• Vendor viability. Whether the company will still exist, and stay independent, in three years.

What to remember

• Find by need, then place by five questions. Built for, legal sources, who builds, how you pay, own copy. They work for any tool, including the next one.
• Specialisation is not intelligence. It buys a fitted workflow, data or experienced people. It pays when the task is narrow and frequent.
• “Legal” on the label does not mean legal sources. Ask which database an answer comes from, and check the citation in the official source.
• AI moves to where your documents are. Check what your current systems already include before you buy a new tool.
• For a product, ask what you can take with you, not whether it is open. Almost every product is the vendor's copy.

Next, lesson 4.6 shows how to choose between the tools on your shortlist.