G

Gist

Every part of the workspace runs on a model. The document editor inside Word and the web app drafts, reviews and redlines. AI retrieval across a firm's files answers questions from its own documents, and skills turn a plain language description plus a firm's templates and examples into a reusable drafting or review routine. Custom agents built by Gist's deployment team run a team's own process end to end, from a first draft of a share purchase agreement to a redline of an incoming NDA.

The workspace keeps a cumulative memory of each team's standards, preferred language and past decisions. Each task is routed to a Claude, GPT or Gemini model chosen for quality, latency and cost, and the platform terms describe the product as an AI powered legal operating system for research, drafting and review.

Answers are grounded in the customer's own documents, and every answer links back to the document and the passage it came from, so a lawyer can check it before relying on it. Retrieval reaches clauses, precedents and matter details across the firm's files, and skills draw on past work and the team's standards.

Output may contain errors, hallucinations or inaccuracies and may be incomplete, by Gist's own account, and evaluating each output is left to the customer. No accuracy figure, test set or error rate is published, and nothing describes research against primary law or a check of cited authority.

Under its terms, Gist is a review, research and drafting tool whose output the customer evaluates in each case, with human review as needed, and never relies on without human judgment. For autonomous or semi autonomous agents, the customer sets each agent's scope of actions, its permissions and any human approval requirements. Users who are not licensed lawyers must have output reviewed or confirmed by one.

Drafting and review happen in an editor inside Word or the browser, and the deployment team maintains each custom agent's prompts and performance. Default approval points, the actions an agent may take on its own and what happens after an agent step goes wrong are not published.

No customer is named, and no deployment, outcome or usage figure is published. Gist lists the kinds of work its agents already run for law firms and in house teams without naming a firm or measuring a result. The examples are drafting from a firm's precedent, answering USPTO office actions, triaging NDAs and filling security questionnaires. Its commitment to deliver a custom agent within two weeks describes Gist's own process rather than a customer's results.

Section 4.7 of the platform terms bars third party model providers from training on customer content, and Gist opts out of every training option each provider offers. Zero data retention is enforced across every foundation model Gist routes to, and free users get the same protection as paying ones. Customer content is confidential information under section 7, and personal data processed for the customer is open only to employees bound by confidentiality.

For its own purposes, Gist may use deidentified and aggregated usage data to improve and develop its products, and that data includes anonymized content and anonymized elements derived from customer content. It sits beside a public statement that Gist never trains on customer data. Privilege and work product are not addressed in the terms, though the trust center calls privilege a design constraint, and nothing describes separation between matters or users inside a customer.

The terms call Gist a multipurpose AI legal assistant intended for legal professionals that does not replace professional judgment. They state that its output is not legal advice, creates no attorney client relationship, and may contain errors or hallucinations that the customer must evaluate case by case.

A user who is not a licensed lawyer, and does not work under a lawyer's direction or supervision, must have every output reviewed or confirmed by a licensed lawyer. Jurisdiction limits are not stated, and nothing describes how the product supports a lawyer's competence or supervision duties beyond those terms.

No AI policy, governance framework, accountable owner or testing program is published. Model selection between Claude, GPT and Gemini is described in terms of quality, latency and cost, and the deployment team maintains prompts and performance as models change. No method, test result or review of uneven output across matter types, practice areas or languages is disclosed.

Personal data handled for a customer is used only to provide the service or as the customer instructs, and only staff bound by confidentiality can reach it. Gist publishes its subprocessors and gives advance notice of changes, with three days to object after publication. Customers can retrieve their content for 30 days after a subscription ends, after which Gist deletes it securely, and model providers keep nothing under zero data retention.

Breaches are reported without undue delay, followed by an investigation and steps to prevent recurrence. Incident response policies and tested continuity and recovery plans sit among the trust center's controls, along with encryption in transit and at rest, restricted key access, MFA for remote access and periodic access reviews. How long content is held during an active subscription is not stated.

The platform terms set out who bears what. Each party's total liability is capped at the fees paid or payable for the services behind the claim in the preceding 12 months, and lost profits, lost data and indirect losses are excluded. For a breach of Gist's data protection obligations, and for its privacy indemnity, the cap rises to three times those fees. Fraud, willful misconduct, intellectual property misuse and indemnities sit outside the general cap.

Customers are defended and indemnified against claims that the platform, its documentation or its output infringe a third party's intellectual property, excluding customer content, misuse, trial services and combinations Gist did not supply. Gist also indemnifies for its own breach of privacy obligations. The service is warranted to conform to its documentation, with a 30 day cure period and a prorated refund if the fix fails.

Beyond that, the platform and output are provided as is, and output is acknowledged to contain possible errors and hallucinations.

Lawyers keep working where they already do. The document editor runs as an add in inside Microsoft Word and in the web app with the same context, and Gist in Outlook brings the workspace into email, so drafting and review need no copying between windows. AI retrieval across a firm's files works through several connections, which are not named, and custom plans add integrations built by Gist's engineers into the firm's existing systems.

No document management system such as iManage or NetDocuments, and no practice management system, is named as a connection, and what syncs in each direction is not described.

The platform runs as a cloud service hosted with third party providers, with Google, Microsoft and Amazon Web Services named as cloud subprocessors. Custom plans add private cloud deployment.

Hosting regions are not stated, and the website privacy policy says information may be stored and processed in jurisdictions around the world. Data residency is named as a design constraint in the trust center, without a list of the regions on offer or a split between where processing happens and where data is stored.

A SOC 2 Type II report audited by EY covers the platform, and the same protections apply on every tier, including free. The trust center at trust.gist.legal sets out 56 controls across product security, access management, continuity, data security, organization security and endpoints. They include penetration testing, MFA on remote access, access reviews, tested disaster recovery plans and device management. The report itself is requested through the trust center, and the audit period and the systems in scope are not shown.

Each task is routed between Claude, GPT and Gemini models, picked for quality, latency and cost. The subprocessor list names Anthropic and OpenAI as AI providers and Google and Amazon Web Services as AI and cloud providers. Zero data retention applies across every foundation model Gist routes to, and the terms bar the providers from training on customer content.

Changes to the subprocessor list come with advance notice and three days to object, which covers a new model provider. Which model versions are in use, and the regions where they run, are not stated.

Prices are published in USD for the self serve platform. Professional costs $80 per seat a month and Team $180 per seat a month, each billed yearly; monthly billing is also offered, and yearly saves 20%. Professional comes with all AI models, the web editor and Word add in, drafting, research, review and redlining, retrieval across files, SOC 2 controls and SSO. Team adds a shared workspace, central billing and user management, and dedicated onboarding and training.

Every plan starts with a 14 day free trial. Custom plans with bespoke agents, integrations, private cloud deployment, a dedicated solutions attorney and ongoing engineering support are priced on contact.

Buyers are law firms, from boutiques to full service firms, and in house legal teams. For firms, the examples include drafting a Series A share purchase agreement from the firm's own precedent and responding to a USPTO office action along the firm's decision tree. Reviewing an incoming contract against the firm's standards and answering status questions from a matter file are among them too.

For in house teams they include triaging NDAs and vendor agreements, answering routine business questions from internal policies, filling security questionnaires and tracking obligations and renewal dates. Practice areas, jurisdictions and work the product does not support are not listed.