Avvoka vs Clarilis: how they compare in 2026

Avvoka combines AI assisted template creation with a document assembly and negotiation system. Curate compares clauses across uploaded agreements and proposes a reusable precedent, while the lawyer chooses the language to retain. MassDraft uses AI to identify inconsistent input data before generating many agreements. Beneath these functions, questionnaires, conditional clauses, approved libraries, version history, negotiation controls, and signatures remain usable without a model. The assistant connections bring that governed drafting engine into other tools. Removing AI would remove comparison and preparation capabilities, but the structured document workflow would continue to operate. The product therefore has an AI engine within a broader drafting platform.

Clarilis describes deterministic, template based automation as the gold standard for legal drafting and positions AI Draft as adding to it rather than replacing it. It gives the split: automation takes a draft roughly ninety percent of the way and AI Draft helps with the remaining ten. AI Draft creates new content without changing the content the rules generate, and its FAQ says firms that do not permit generative AI can disable AI Draft in an automation, leaving the rest of the platform unaffected. What remains with it switched off is the whole product: questionnaire driven assembly of complete document suites, hundreds of maintained templates across four practice areas and five jurisdictions, house style formatting, and the managed service behind it.

Curate works from files supplied by the customer, compares versions of each clause side by side, and lets the lawyer select, combine, or rewrite the wording before it enters a precedent. Subsequent document assembly follows approved template rules. The drafting process is therefore source based, though deterministic assembly cannot show that an AI synthesis or the underlying legal language is correct. Avvoka advertises millions of drafted documents without hallucinations without publishing a reproducible evaluation or failure analysis. The product does not supply a legal authority citator. Its source controls concern the firm's own contract language and require professional review.

There is no accuracy figure, evaluation, test set or error rate on the site. The ninety percent figure that recurs is a time saving claim about the automation, not an accuracy measure. AI Draft writes new contract language and has no authority to cite. The vendor lists what its own AI gets wrong. Users must check AI content for cross references, because the AI does not yet match references to other clauses automatically; for definitions used but not defined elsewhere; and for legal and commercial effectiveness. The guidance ends by telling users to always treat AI generated content as a first cut. Grounding is described as matter context and codified practice area knowledge from in house lawyers, and how that grounding works is not described.

Curate separates model assistance from adoption into the firm's precedent. It compares uploaded clauses and can combine wording, but no clause enters the template until the lawyer approves it. After approval, document generation follows the template's rules. Negotiation controls can lock clauses and route work to named people, with comments assigned as tasks and versions preserved. MassDraft adds validation of input data before bulk generation and lets users compare amendments before execution. These controls define both automated actions and review surfaces. Approval of a clause is an explicit boundary for Curate, although the vendor does not publish numerical stopping conditions across all AI functions.

AI Draft writes new content only where a user invokes a named task, and it cannot change the drafting the automation produced. Review is built into the document itself. Every AI generated clause is highlighted in the delivered Word document, and any document containing AI content carries a notice at the top saying that content needs review. The vendor says AI content is new and, unlike automated content, has not been reviewed in advance, and lists what to check: cross references, use of definitions, and legal and commercial effectiveness. It calls AI generated content a first cut that still needs expert review. A firm that does not permit generative AI can turn the layer off and keep the automation. Because the marking sits inside the document, it survives export, which a review state inside the platform does not.

Named users describe Avvoka in production. Maddocks identifies its replacement of a legacy platform and reports improved stability and easier adaptation as document requirements change. Ropes & Gray describes adoption and workflow improvements, while Fried Frank's knowledge leadership discusses turning past deal language into approved standards through Curate. The homepage also advertises drafting eight times faster and adoption by a quarter of the AmLaw 100. Those aggregate figures are not accompanied by a defined measurement period or method. The attributed accounts establish actual use and customer context, but do not isolate an independently measured benefit from the AI functions.

Two customers are quoted with organizations attached. Miri Stickland, Head of Knowledge at Forsters, speaks to the risk management benefit of updates being made quickly and consistently behind the scenes. A Head of Knowledge Management at TLT, named by role only, says the solution exceeded expectations on delivery. Both link to individual customer pages, and CMS is named in a vendor blog post about faster drafting for its Scottish real estate team. Separately, the platform material says the system cuts the time to produce document suites by around ninety percent, and that firms draft suites ninety percent faster, with no firm, date or method attached to either figure. So the named customers carry no measurement and the measurement carries no named customer. A customers section and a case studies section sit in the navigation.

Documents are encrypted at rest, connections use HTTPS, and administrators can require a second authentication factor or single sign on. Curate and the assistant drafting offering promise that customer documents will not train a shared model. That wording does not expressly prohibit all forms of private model adaptation. The legal advice disclaimer also states that information supplied to Avvoka does not attract legal advice privilege through a relationship with the software provider. The assistant offering claims that permissions and ethical walls carry over, but does not explain enforcement. No readable customer agreement provides a complete confidentiality, retention, and model provider commitment.

On AI, neither user inputs nor AI responses are used to train the models, processing runs through zero data retention endpoints, and data is processed in region. On the platform, encryption covers data at rest and in transit, and protected links expire and use 256 bit AES. All requests are logged and verified, including IP addresses, and user level checks restrict access to particular document suites and drafts. SAML 2.0 single sign on and optional IP lockdown are offered, and development, testing and production environments are separate. All staff pass criminal record and eligibility checks, access follows least privilege and is recorded and auditable, and system development is not outsourced. No privilege or work product treatment appears anywhere, on a platform holding transactional drafting for named clients. No customer agreement, master services agreement or data processing addendum is published, so every commitment above sits on a product page rather than in a contract.

A full advice line statement is published on a dedicated page, with no engagement with the professional rules. The disclaimer says Avvoka is not a firm of solicitors, does not practice law and does not give legal advice. It is not the user's solicitor or a substitute for a solicitor's advice, and has no solicitor client relationship with the user. A user who enters into a document created on the platform does so at their own risk and represents themselves. It also lists what the product does not do: it does not review information for legal accuracy or sufficiency, draw legal conclusions or give opinions on the documents a user creates, or apply the law to the user's circumstances. It says information provided is not covered by legal advice privilege. No Solicitors Regulation Authority guidance, bar rule or ethics opinion is named, and nothing addresses a firm's supervision or competence duties when generated drafting reaches a client. The page opens as terms for use of the website, though its substance plainly covers documents created on the platform.

The vendor says AI Draft does not replace professional legal judgment, that AI generated content should always be treated as a first cut that still needs expert review, and that automated content is carefully reviewed in advance while AI content is new and has not been. Nothing says Clarilis is not a law firm or that its output is not legal advice. No jurisdiction limit is drawn, although automated content spans England and Wales, Scotland, Ireland, the United States, Canada and Asia Pacific, and nothing addresses a firm's supervision or competence duties. There is no terms of service on the site where such a statement could sit, so the position rests entirely on a product FAQ and a blog post.

Information security controls and an information security management policy are set out, with no corresponding AI governance framework. The materials do not identify a model risk owner, release approval process, evaluation regime, or testing of differences across languages, legal contexts, and document types. Customer approval in Curate gives a lawyer control over the content of a template. It does not explain the vendor's testing of the models that propose that content. ISO 27001 certification concerns information security and is not a statement that model fairness, legal reliability, or bias has been independently assessed. No AI management certification is identified.

Clarilis says it has signed the Litig AI Transparency Charter, links to it, and describes it as setting out commitments for organizations developing and providing legal AI, promoting responsible innovation while safeguarding ethical and professional standards. The security page adds a responsible AI section with four commitments: clear indicators separating AI from rules based drafting, processing in region through zero data retention endpoints, no use of data for model training, and respect for third party intellectual property rights. No management system such as ISO 42001 or independent audit of AI governance is published. No accountable owner or function is named, no evaluation regime before release or results are described, and nothing covers uneven output across practice areas or jurisdictions. The ISO 27001 certification concerns information security, not AI governance.

Security controls include AES 256 encryption at rest with a separate initialization vector per document and a rotating master key; HTTPS for all connections, including the API; BCrypt password hashing, with passwords kept out of logs and rotation rules available; enforceable two factor authentication; SSO only mode with automatic removal of leavers; real time replication of production databases with hot standby failover; and encrypted backups. Annual black box penetration testing uses only CREST approved providers, with the latest scorecard available to clients on request. There is a published vulnerability reporting process with a PGP key, and the build pipeline is continuously tested for CSRF, XSS and SQL injection. A breach procedure commits to identifying compromised information and notifying affected individuals and the bodies the law requires. No retention period is stated; the GDPR policy says only that data is kept when necessary and deleted on a data subject's request. No subprocessor list goes beyond the named infrastructure providers.

Published controls cover encryption at rest and in transit, protected links that expire and use 256 bit AES, servers behind a firewall with all access encrypted, and separate development, testing and production environments. Data sits on encrypted file systems backed up at least daily, with horizontal scaling and resource monitoring. All requests are logged and verified, including IP addresses, and user level checks limit access to particular document suites and drafts. SAML 2.0 single sign on comes with optional IP lockdown and configurable lockout. All staff pass criminal record and eligibility checks before hiring, access follows least privilege and is recorded and auditable, and development is not outsourced. Hosting is named as AWS and Azure, and the AI layer uses zero data retention endpoints in region. Regular penetration testing is stated. No retention period is given: the only statement is that the customer has complete control over storage and deletion of data, with no period or process described. No incident notification commitment appears, and there is no agreement for one to sit in.

Nothing published says who bears the loss when the product is wrong. The published legal set is a GDPR policy, a privacy policy, a legal advice disclaimer, a signing disclaimer and a security page. There is no terms of service, customer agreement, subscription agreement or service level agreement, so there is no indemnity, liability cap, warranty, warranty disclaimer or insurance statement to read. The legal advice disclaimer says a user who enters into a document does so at their own risk, but that concerns advice, not loss. It does not cover a defective template, a clause assembled wrongly, a negotiation lock that failed, or a generated draft that departed from the firm's standards. No standard limitation clause is published either.

Nothing published addresses who bears the loss when the product is wrong. The site footer carries two entries, a privacy policy and a sitemap. There is no terms of service, master services agreement, customer agreement or data processing addendum anywhere on the site, so no indemnity, liability cap, warranty position, warranty disclaimer, service level commitment or insurance statement is published. The product produces complete suites of transactional documents through a managed service, and nothing says who carries a defect in a maintained template. No limitation clause of any kind is published.

Assistant connections expose templates, approved clauses, amendments, and document export through other working environments. Claude and Microsoft 365 Copilot are listed as live, Harvey as early access, and Legora as a planned beta in the fourth quarter of 2026. Curate names an iManage integration, while API and MCP access support connections to business systems. MassDraft describes Docusign envelope preparation, signature status tracking, and combining signed counterparts for download. These sources explain actions beyond a partner logo. They do not publish complete field mappings, configuration steps, or permission synchronization behavior across every connector, so implementation still needs discussion with the vendor.

No integration into the systems a firm runs is published. The platform is web based and delivers finished drafts in Word in the firm's house style, which is a format and a delivery convention rather than a connection. Two named third parties appear and both are identity rather than practice systems: SAML 2.0 single sign on for authentication, and Certivox M-PIN offered as an optional additional security element. No document management system is named anywhere, so nothing addresses iManage or NetDocuments, and no matter management, ebilling, CRM or electronic signature counterparty appears. No API, developer documentation or integrations page exists in the navigation, which carries platform, AI Draft, security, managed service, partners and working globally.

Avvoka is delivered as a hosted platform. Its general security materials name infrastructure in the United Kingdom, Europe, the United States, and Australia through AWS, OVH, and Azure. Curate and the assistant drafting offering separately specify European Union residency. The GDPR policy describes an older arrangement with a United Kingdom application server and French backup, so that policy alone does not explain the full current regional offering. The materials do not map every model operation to a processing location or define single tenant deployment options. A customer needs the selected product and contractual region to establish the boundary for its own content.

The AI Draft page says data is hosted in the customer's region via Azure, and the security page says the same for the AI layer: data is processed in region using zero data retention endpoints. The platform serves firms across England and Wales, Scotland, Ireland, the United States, Canada and Asia Pacific, and as a whole is hosted on Amazon Web Services and Microsoft Azure. Nothing says whether the platform is single tenant or multitenant, no dedicated or isolated instance is described at any tier, and no on premises or self hosted option is mentioned. The nearest thing to a separation statement is a permission control: user level checks restrict access to particular document suites and drafts. That governs users within a customer, not the boundary between customers.

A named standard is backed by a stated history and a published document, without the certifier or scope. Avvoka says it has been ISO 27001 certified continuously since October 2017, with annual external audits, and it publishes its signed ISMS policy as a PDF anyone can download. A period, an audit cycle and a readable governing document go well beyond a badge. Penetration testing is annual, black box and done only by CREST approved providers, with the latest scorecard available to clients on request. Avvoka gives no certification body, certificate number, scope or statement of applicability, so a buyer cannot tell which systems and locations the certificate covers. There is no trust portal.

The security page says Clarilis is ISO/IEC 27001:2022 certified by BSI across the entire company and gives the certificate number, IS 677941. That covers the standard, the current revision, the certifying body and a company wide scope, plus an identifier that can be checked against the certifier's register. Regular penetration testing is stated separately on the AI Draft page. No certification date, expiry or surveillance cycle is published. No report, statement of applicability or summary is published, and there is no trust portal. No other certification is claimed.

AI comparison, synthesis, and validation functions are described without the model versions or providers that run them. Claude, Copilot, Harvey, and Legora are named as assistant connections, not a complete account of Avvoka's internal model stack. Naming the infrastructure providers AWS, OVH, and Azure likewise leaves open which models receive document content. Curate limits its input to uploaded files and promises no training of a shared model, but no public register connects each AI feature to its processor and retention terms. A commitment to notify customers of changes to that underlying supply chain is also not published.

In the AI Draft FAQ, the vendor says it prefers OpenAI's model, hosted in Azure. The AI Draft page adds that data is hosted in the customer's region via Azure and processed through zero data retention endpoints. The vendor also says it uses a range of models depending on the drafting task and configures them with practice area knowledge from its in house lawyers, so the supply chain is plural. Its FAQ says OpenAI's model without naming which model or version, and adds that Clarilis is always exploring the latest models to ensure the best outputs. That says the set will change, and no commitment to notify customers when it does is published. No subprocessor register is published, though AWS and Azure are named as platform infrastructure on the security page.

No pricing information is published at any level, including the unit of charge. The navigation has features, solutions, resources and company sections only, and the footer has no pricing entry. Every commercial route on the site is a demo booking, and there is no self service signup. Nothing says whether the platform is licensed per user, template, document, matter or firm. No tier names, minimum commitment or term, or implementation or template building cost are given, although building automated templates from precedents is a delivery project. No customer agreement is published either, so no payment terms show the charging structure.

No pricing information is published at any level, including the unit of charge. There is no pricing page in the navigation (what we do, solutions, customers, about us and resources) or in the footer, which carries only a privacy policy and a sitemap. Every commercial route leads to the same call to action, book a demo. Nothing says whether the platform is licensed per user, template, document suite, matter or firm. Nothing addresses how the managed service is charged; the vendor's own professional support lawyers build and maintain the automation, so implementation is a service engagement rather than a setup task. No tier names, minimum commitment or term length appears. No terms of service or master agreement is published, so no payment provisions are published either.

Avvoka serves law firms and corporate legal teams, with distinct workflows for template authors, document users, and managers. Current examples include mergers and acquisitions, real estate, commercial agreements, and procurement. A share purchase agreement can be generated alongside disclosure letters and related papers from the same transaction facts, while MassDraft addresses large repapering exercises. Curate helps legal teams turn prior deals into approved precedents without beginning from an empty library. The product does not publish a complete jurisdiction or language matrix, or define which practices require unsupported content. Customers remain responsible for the legal suitability of the clauses and rules they approve.

Four practice areas have their own product lines with counts and named content. Real estate has over seventy automated templates for commercial investment and letting in England and Wales, plus more than fifty Property Standardization Group documents for Scotland. Corporate has nearly two hundred documents covering share purchase agreements, early stage investments, share reorganizations and business purchase agreements. Banking and finance covers leveraged and real estate finance facilities, simplified lending, security documents, legal opinions and APLMA based agreements for Asia Pacific. Private client covers wills and estate planning. Engagement letters and in house legal are addressed separately. Jurisdictional reach is stated product by product, with automated suites for Ireland, the United States and Canada, the Canadian content based on CVCA model forms. Nothing says which practice areas or jurisdictions are not served, no firm size range is given, and no document type is named as one the platform handles poorly.