BRYTER vs Neota Logic: how they compare in 2026
BRYTER and Neota Logic are both no code platforms on which legal teams build their own automation, and both now run AI steps inside those workflows. BRYTER is stronger on its contract. Its data processing addendum treats all customer data as potentially covered by professional secrecy, its agreement bars training on customer content, and it publishes per user prices from 99 euros a month. Neota Logic is stronger on control over the model itself. The customer connects its own model through its own Amazon Bedrock or Azure account, and no model runs by default. A confidence score can route any AI step to a named reviewer. Every call is logged in a trail the customer can export, recording the prompt and response, the model and cost, and the approver. BRYTER's own documents disagree on where data is processed, placing it in the EU in one schedule and in United States data centers in another.
Which one fits
Choose BRYTER if
- Your firm is bound by professional secrecy. BRYTER's data processing addendum treats all data as potentially subject to professional secrecy, limits its own knowledge of secrets to what is strictly necessary, requires subprocessors to be barred from onward disclosure, and requires its staff to commit in writing.
- You want to price AI before a sales call. BEAMON AI costs 99, 149 or 199 euros per user a month on annual plans, with monthly rates shown alongside, a self serve free trial, and an agreement that caps annual price rises at 10 percent.
- You practice in Austria or Germany. BRYTER states it has completed the Austrian bar association's checklist for AI providers, maps its AI suite to named Austrian professional rules, publishes supplementary terms for professional secrecy holders, and integrates with the Otto Schmidt legal database.
Choose Neota Logic if
- You want AI to run on your own model account. Neota Logic's AI Task Node calls a model your administrators have approved, through your own Amazon Bedrock, Azure or other provider account, and no model is invoked by default.
- You want to set when AI output goes to a person. Each AI step captures a confidence score as a workflow variable, a decision point branches on a threshold you set, and review nodes require sign off before a process continues or output reaches a client.
- You need a record you could produce to a regulator or court. Neota Logic logs each AI call against its session and matter with the prompt version, the clause and citation relied on, the model, tokens, cost, confidence and approver, in an exportable trail.
At a glance
All 15 axes, side by side
The same grid applied to every vendor in the index, graded from public sources. Hover a grade to see what the letter means on that axis.
AI Centrality
How much of the product is actually AI. Whether the machine learning is the mechanism the buyer is paying for or a feature layered onto conventional software, and whether the vendor is specific about which is which.
AI drives a core capability on a platform that would still work without it. BRYTER's own formula, on every page, is the power of AI with the precision of workflows, and it names two halves. BRYTER Workflows is a rules based automation platform, built without code, that was the whole company from 2018 and still sells on its own. Without the AI it remains a complete product. BEAMON AI is the AI half and is substantial: three products in Assist, Extract and a Word plugin, its own domain, price list and free trial, and its own place in the agreement's definition of the Software. BEAMON AI can be bought alone at published per user rates, so a buyer can get the AI without the workflow platform. But BRYTER is both halves: a rules engine with an AI suite alongside it, and increasingly running through it.
The models drive a core capability of a platform that would still work without them. The AI Task Node calls a model the customer has approved to extract clauses, classify requests, summarize and draft inside an otherwise deterministic workflow. Neota names contract review, matter triage and document summarization as its AI uses, and an MCP connection in preview lets AI assistants call Neota applications. Underneath is the platform for process, decision and document automation without code that the company has sold since 2010, which runs without any model.
Citation Accuracy and Hallucination Disclosure
Whether the vendor publishes measured accuracy on citations and assertions, grounds output to primary sources, and says plainly what its system does when it does not know. Legal has a documented public record of fabricated citations reaching filed briefs, so an untested claim of accuracy is not evidence.
How answers are grounded is set out, but BRYTER gives no accuracy figure. Hallucination control and references are named as features. Summaries are described as linked to their sources, with citations kept and detail selectable. Assist is described as answering from the customer's own knowledge bank and playbooks. The retrieval sources are named: a web search processor in the subprocessor list and a direct API link to the Otto Schmidt legal database on the top tier. The agreement adds a frank disclosure. Accuracy may suffer when web search or access to legal sources is not enabled, the service is then more likely to be wrong, and BRYTER recommends turning them on. That tells a buyer when the product gets worse. No accuracy figure, error rate, test set or evaluation is published. The only comparison is unmeasured marketing, that its AI beats generic tools like ChatGPT or Copilot on quality and accuracy. The agreement is far less confident. It says BRYTER does not guarantee Output is accurate, complete, original, fair, unbiased or free of errors, and where the two differ the agreement governs.
Grounding is documented, with each AI step traced to its source and checked by rules, but no accuracy figures are published. The AI Solutions page shows an extraction step that captures the clause and citation it relied on and attaches a confidence score. It validates the output against the customer's deterministic rules and escalates to a person when a check fails. It also describes the failure the design guards against: a fluent, confidently wrong answer from an ungoverned model. No accuracy figures, test results or failure rates for the AI Task Node itself are published.
Autonomy and Oversight Model
What the system decides on its own, what a lawyer must approve, and whether the vendor documents where the review point sits. A tool that drafts under review and a tool that files without one are different products and different risks.
Human review of the AI's work is a contractual commitment at BRYTER, which also gives real control over which models run. The agreement says the customer must review and validate all results before relying on them for business, legal or other decisions. Using Output without proper human oversight, especially for decisions or compliance, is at the customer's own risk, and Output must be verified before any use. Another clause bars the customer from misleading anyone that Output was written by a person. The best control is over models. An administrator chooses which language models are enabled for the tenant and can disable any at any time. Within that set, the author picks the model for each input, and both choices count as instructions under the data processing addendum. By default only one model is enabled, and no customer data reaches any other provider until an administrator turns it on. Access is restricted by role, actions are logged and changes can be traced. What is missing: no published limit says what a given tier's output may not be used for. The commitment is a general promise of human review, plus control over which models run.
What the AI runs alone, what constrains it and how a person checks it are all published. The AI Solutions page describes the AI Task Node's modes (extract, classify, draft), each scored for confidence and logged. A decision gateway branches on AI confidence against a threshold the customer sets. Human review nodes must approve before the process continues or output reaches a client, and a drafted output is always checked downstream. Neota's FAQ says the customer decides where review points sit and what happens when a check fails: routing to a named person, holding the matter or asking the requester for more. The August 2026 release says no model is ever invoked by default.
Operational and Outcome Evidence
Named, dated evidence that the product works in production at real firms or legal departments. Case studies with figures and identified customers count. Unattributed testimonials and launch announcements do not.
Named customers span three buyer types: global law firms (Linklaters, Ashurst, Hausfeld), professional services firms (Deloitte, PwC, KPMG) and corporate legal departments (McDonald's, ING Bank, Rakuten and Telefonica). A published customer story covers Telefonica's legal team, and one testimonial carries a named person with title and firm, a counsel and AI expert at REIUS. No figure is tied to a named customer on the product pages, and nothing is dated. The benefit claims are generic, such as saving hundreds of hours reviewing contracts, with no basis, population or period. Nothing measures BEAMON AI apart from the workflow platform. BRYTER also publishes a customer stories library and a guides and reports section.
Named deployments have substance but no measured result. Six case studies cover Harneys, Fujitsu, Thommessen, Simonsen Vogt Wiig, Grampians Health and an unnamed firm's FCPA subscription service. Harneys built an application for launching funds, after which the firm introduced fixed fees and reports a higher pitch win rate. Fujitsu handles NDAs, DPAs and other agreements on a self service basis, Thommessen runs matter onboarding integrated with Salesforce, and Simonsen Vogt Wiig built a public guidance tool that won an award. Grampians Health handles freedom of information requests, with letter readability improved from grade 8 to 12 to grade 3 to 4. None gives a before and after figure with a method, and the readability change reflects the customer's own rewriting. The home page claims more than $100 million of verified value with no basis. None of the case studies concerns the AI features.
Privilege and Confidentiality Posture
How client confidences are handled: attorney client privilege and work product treatment, segregation of one client matter from another, whether client data trains any model, and what the vendor commits to in writing rather than in marketing.
BRYTER addresses privilege and professional secrecy directly, in documents a buyer can read before signing. The data processing addendum has a section on BRYTER's duty to keep professional secrets, which applies wherever the customer has such duties. BRYTER treats all data as potentially subject to professional secrecy unless clearly determined otherwise. It may learn professional secrets only as strictly necessary, and learning expressly includes any technical or organizational access, whether or not a person reviews it. It must keep them strictly confidential. It may pass them to subprocessors only where they are contractually barred from passing them on and must flow the duty down, and its own staff must promise in writing not to disclose them. BRYTER also states compliance with German professional rules and the criminal law on professional secrecy, and says it upholds attorney client privilege. Training is banned by contract. Retention is set by data category, with deletion or return at the customer's choice and a limited backup window. Model providers are named with processing locations, and the default model is stated to run in the EU. Separation is documented through tenant separation and a role concept. One point weighs the other way. The master agreement makes the customer solely responsible for ensuring any Input complies with privilege and professional secrecy duties, so BRYTER protects the material but disclaims the decision to submit it.
The confidentiality position is structural and published, but no customer contract is published and privilege is not addressed. Neota's FAQ says customer data is not used to train models because the customer connects its own model under its own provider agreement inside its own AWS Bedrock or Azure boundary, so existing provider protections stay in place. Application data is encrypted at all times and deleted when a session ends unless the site owner directs otherwise. Access to applications is set by the customer down to who can see a matter. No customer agreement is published, and privilege and work product are not addressed.
UPL and Professional Responsibility Posture
Whether the vendor is clear that it supplies a tool rather than legal advice, who its audience is, and how it addresses unauthorized practice of law, competence and supervision duties, and jurisdiction limits. ABA Formal Opinion 512 is the reference point. Where the advice line is not the duty a product raises, the axis is read through the nearest professional duty it does raise: judicial conduct rules and the reviewing duty for products sold only to courts, and the duty to bill for time actually spent for products that draft time entries.
Output is not legal advice under BRYTER's contract, which addresses supervision but sets no jurisdiction limits. The agreement says Output is not legal advice, that BRYTER is not a law firm, does not practice law and gives no legal advice, and bears no legal responsibility for Output or anything derived from it. The usage rules add two less common provisions: a bar on misleading anyone that Output was written by a person, and a bar on giving tailored financial advice without a qualified person reviewing it. The agreement's review and validation duties cover supervision of legal output. BRYTER also engages professional conduct in at least two countries, mapping its AI suite to named Austrian provisions and stating compliance with German professional rules and the criminal law on professional secrecy. Missing: no jurisdiction limits. Nothing tells a buyer where the product may or may not be used. That matters for a company contracting under German law, selling in the US through a separate entity, in at least three language markets, with added terms for Austrian and Swiss professional secrecy holders. Who may operate the tool is not addressed beyond the license terms Authorized User and Author.
Neota states a real position on what the tool is, without guidance for the advice tools customers build for their clients. The Terms of Use tell users to have a qualified person review outputs before relying on them. Neota's FAQ says the model does the language work while the customer's rules decide the outcome, with the expert knowledge coming from the legal team rather than from Neota. Customers use the platform to publish advice applications to their own clients and to the public, such as Simonsen Vogt Wiig's free guidance tool. Nothing addresses disclosure, supervision or jurisdiction for those applications.
AI Governance and Bias Disclosure
Published governance over model behavior: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.
A governance framework is written into BRYTER's contract, without test results or a named owner. The master agreement has a dedicated AI Terms section for every AI feature. It bans training on customer content, governs model selection and change, requires human oversight, and sets acceptable use rules. Those rules bar hateful, harassing or violent content, exploiting children, encouraging self harm, misusing personal data, malware, high risk economic harm, health advice, political campaigning and lobbying, and uses that breach privacy or confidentiality. Bias is addressed in the agreement itself. The customer acknowledges that Output may reflect or amplify biases in training data and agrees to verify and validate all Output before use. Around this sits company governance: ISO 27001:2022 certification, a data protection management system reporting to senior management, records of processing under Article 30, privacy by design training for product teams, and internal and external audits. Missing: no one is named as accountable for AI, and nothing describes testing before a model or feature ships. No test results are published, and the bias clause is a warning, not a mitigation.
There is a governance framework with real mechanisms, and it sits on the customer side. Models are approved centrally by the customer's administrators, and no model runs by default. Every call is logged with prompt, response, model, cost and approver, and outputs are validated against the customer's rules, with human approval where the customer sets it. Neota sells this governance as its product, and the same controls apply to Neota's own AI Task Node. Nothing is published on how Neota tests the node before release, who is accountable internally, or uneven performance.
AI Safety and Data Stewardship
Retention, deletion, access control, and what happens to prompts and documents after they are processed. Whether the vendor states its subprocessors and its incident practice, or leaves the buyer to assume.
Every element of data stewardship is set out, including a subprocessor table. Retention is set by data category: security and audit logs up to ninety days, contact, credential and customer content until termination and then deletion, and backups for thirty days after. On termination BRYTER deletes, or returns and deletes, all personal data and processing results at the customer's choice. There is also a right to switch and port data under the EU Data Act, with a documented process. Access control is documented under twelve headings. They cover physical access, user and data access, separation, pseudonymization and encryption, input and order control, transmission, availability, resilience, security management, purpose limitation and data protection management. Subprocessors are named with service, company location and server location. Changes need prior written notice with a twenty business day objection right. Incidents are notified without undue delay, and where feasible within 72 hours, with four required contents. Business continuity, incident response and disaster recovery plans are tested. Two frank disclosures. The addendum discloses monitoring of AI processing, described as LLM observability. And the provider of the chosen language model screens inputs and outputs automatically against its own usage terms, with human review of flagged content case by case.
Most of the ground is covered, except a subprocessor list and an incident commitment. The Security page states AES-256 encryption in transit and at rest, role based access with single sign on, regular third party penetration testing and hosting on AWS or Microsoft Azure. The Privacy Policy says application data is encrypted, deleted at the end of a session unless the site owner directs retention, and stored in the country or region the site owner chooses. AI calls run through the customer's own model account. No subprocessor list or incident notification timeline is published, and a security pack is available on request.
AI Liability and Recourse
What the vendor stands behind contractually when its output is wrong. Indemnities, caps, carve outs, insurance, and whether any of it is published or only reachable through a negotiated agreement.
BRYTER publishes real warranties, not only exclusions, but carves the AI out of the strongest ones. It warrants that it can grant the rights, that the Software will be provided with reasonable skill and care and in line with law, that it will materially match the Documentation and the Order, and that functionality will not materially decrease during the term. The remedy is prompt correction or another way to do the task. There is an uptime commitment backed by a support and maintenance appendix, with response targets of 48 hours standard and 24 hours on the top tier. Liability follows German law. It is unlimited for intent, gross negligence and injury to life, body or health. Slight negligence is limited to breaches of material obligations and capped at the fees for the current subscription period. The catch is the AI. The agreement says support and maintenance apply to the AI Service only to the extent it may have downtime, interruptions and errors, with no guaranteed response or fix time. It also says BRYTER does not warrant the availability, accuracy or completeness of results from credit based features. So the uptime promise covers the platform, not the models. No indemnity to the customer was found, no insurance is published, and the only indemnities run the other way.
Liability is addressed only through a standard limitation. The Terms of Use (1 June 2026) provide the services as is, tell users to have a qualified person review outputs before relying on them, and exclude indirect and consequential damages. They limit aggregate liability as set out in the customer's agreement, which is not published. No indemnity or position on incorrect AI output is published.
Practice Systems Integration Depth
How deeply the product reaches into the systems legal work already lives in: document management such as iManage and NetDocuments, Word and Outlook, contract lifecycle management, matter management, e-billing, and court filing systems.
Document and knowledge systems: iManage and SharePoint, connecting repositories so precedents appear in context. Business systems: Salesforce and DocuSign. Collaboration: Microsoft Teams, Slack and email as intake channels. Authoring: a Word plugin for drafting, editing and redlining. Publishers and data: a direct API link to the Otto Schmidt legal database, Haufe Lexware, Winsolvenz insolvency software via Septeo, and Handelsregister company data via Company.info. Language: DeepL, also in the subprocessor list with its processing locations. Platform: BRYTER Connect and an open API, with published API terms, a license to build on it, software kits and code samples. Modules can be triggered from another system and data exported to an internal dashboard. The pricing table states which integrations need the customer's own third party subscription and which are paid extras. BRYTER runs a help center; outside it, no public API reference or connector list is published, and for most connectors nothing says what data moves in which direction or what a firm must set up.
Connections into legal systems are named with what they do, but there is no documentation for implementers. The Integrations page and FAQ list prebuilt connectors for iManage, NetDocuments, SharePoint and Box (store, retrieve and file documents), Microsoft 365 and a Word Toolbar, and DocuSign and Adobe Sign (route for signature). They also list Salesforce, Dynamics 365 and HubSpot (push and pull data), Slack, Teams, Gmail and Outlook, PostgreSQL, MS SQL, Power BI and Tableau. Inbound and outbound REST APIs are offered with a sample call, and an MCP server is in preview. What syncs in which direction and what a customer must configure are not documented publicly, and training runs through Neota University.
Deployment Model and Data Residency
Where the software runs and where the data sits. Multi tenant cloud, single tenant, private deployment, on premises, and whether region of residence is a published option or an enterprise conversation.
BRYTER publishes detailed residency information, but its own documents contradict each other twice on the key point. The pricing table states EU hosting for every tier. The subprocessor schedule gives a server location for each party: cloud hosting in Frankfurt, the default language model hosted by Microsoft Azure in the EU, and other named locations. A clause says that by default only one model is enabled, processed within the EU, and that no customer data goes to another model provider until an administrator enables it. Tenancy is covered by the Tenant concept in the agreement and by documented separation of customer data into separate systems. So a buyer is told both where data is stored and where AI processing happens. The contradictions are not resolved in BRYTER's favor. First, within the data processing addendum, the subprocessor schedule lists servers in Frankfurt, the EU, Sweden and the Netherlands. But the technical measures schedule says all personal data is stored and processed in US data centers of AWS and Microsoft Azure. Second, the security page says customer data, prompts and outputs never leave the EU and EEA, but the schedule lists an optional model provider and a web search processor with servers in the USA. A buyer cannot reconcile these from what is published.
Deployment options, residency choice and the location of AI processing are stated, but not the regions available or what changes between options. The Security page offers hosting on AWS or Microsoft Azure, and the Pricing page offers cloud, on premises and private cloud. Neota's FAQ says the deployment region is chosen to fit the customer's residency requirements, and the Privacy Policy says site data is stored in the country or region the owner directs. AI runs inside the customer's own Bedrock or Azure boundary. No list of regions is published.
Security Certifications and Trust Center
Independent attestation a buyer can pull without a sales call: SOC 2, ISO 27001, penetration test summaries, a trust center with current reports and named scope rather than a badge image.
SOC 2 Type II and ISO/IEC 27001:2022, with the version year, are claimed on the pricing table against every tier and on the security page. The data processing addendum records internal and external ISO 27001 audits, sets out technical and organizational measures under twelve headings, and names the certifications of its two cloud subprocessors. No auditor, report period or certificate number is published on those pages. The pricing table links to trust.bryter.io, a dead address, while the security page links to the trust center at trust.bryter.com.
A certification is named without scope, date or a published route to the certificate. The Security page states ISO 27001 certification and regular third party penetration testing, and offers a security pack through the contact form. No certificate number, scope, certification body, date or trust center is published.
Model Supply Chain Disclosure
Which models sit underneath, whose they are, where they run, and whether the vendor commits to telling customers when that changes. A legal buyer inherits every dependency it cannot see.
Model providers are named, and customers control which ones run, but BRYTER names no model. The agreement and subprocessor schedule identify Microsoft Azure as the cloud host of the default language model, and Anthropic Ireland as an optional model used only when a customer administrator enables it. A web search processor is named for the AI suite, and the usage restrictions also name OpenAI and Azure OpenAI among the third parties providing the AI Service. Each has a company location and a server location. Change notice is a real process: prior written notice before adding or replacing a subprocessor, a twenty business day objection right on material grounds, and a defined result if the objection stands. Customers control which provider handles what at two levels: administrators enable providers per tenant, and authors choose per input. No model is named anywhere. The pricing table says only that available models are updated continuously. The agreement lets BRYTER add, replace, upgrade or drop models at any time, with no customer claim to any specific model or version. A buyer knows whose infrastructure runs the AI, but not which model is answering.
The model providers are named, where they run is stated, and no model changes without the customer's approval. The Integrations and home pages name OpenAI, Anthropic Claude, Google Gemini, Azure AI Foundry, Amazon Bedrock and models the customer hosts itself. Neota's FAQ says deployment sits inside the customer's own AWS Bedrock or Azure boundary under the customer's own provider agreements, and the audit record names the model on every call. The August 2026 release says models are approved centrally by the customer and none is invoked by default, so the customer controls any change rather than relying on notice from Neota.
Commercial Transparency
Whether a buyer can learn what this costs without entering a sales process: published rates, the unit being charged, what sits behind an enterprise tier, and what implementation adds.
A buyer can price BRYTER's AI suite to the euro without speaking to anyone, and can read the contract terms behind it. Rates are per user per month. Single is 99 euros annual or 118 euros monthly, for solo practitioners without team features. Team is 149 euros annual or 176 euros monthly, adding shared projects, documents and grids with a central prompt library. Pro is 199 euros annual or 234 euros monthly, adding workflow automation, single sign on, integrations and priority support. Both billing options are shown, so the annual discount can be worked out. A comparison table of about eighteen rows covers features, integrations and publishers, and compliance and support. It marks which integrations need the customer's own third party subscription and which are paid extras. Purchase is self serve, with a free trial and sign up, and published terms for buying through Stripe. The agreement also publishes terms buyers usually find later. These include annual invoicing in advance, yearly increases capped at ten percent, ninety days' notice and a right to leave if fees change at renewal, and thirty days' notice for credit rate changes. There is a monthly credit cap with unused credits forfeited, a token limit with notice at eighty percent, thirty day payment terms and a stated late payment interest rate. Two limits. BRYTER Workflows is not priced publicly and is quoted, and premium model access may cost extra under the Order.
The unit and structure are published without the figure. Neota's pricing page says it is priced per solution or use case rather than per user, on a tiered subscription in which the cost per solution falls as more are built. Professional services are sold at a fixed fee. Deployment options include cloud, on premises and private cloud, and a Discovery Sprint with a recoverable upfront investment is the way to start. The FAQ adds that price reflects solutions in scope, users and environments. No rate is published.
Firm and Practice Coverage
Who the product is actually built for. AmLaw, midlaw, small firm and solo, in house departments, government and courts, and which practice areas are supported rather than merely claimed.
Coverage spans segments and tasks, with no limits stated by BRYTER. Law firms and corporate legal teams each have their own solution pages and use cases. The customer list shows all three buyer types BRYTER claims: global law firms, professional services firms and corporate legal departments. Firm size is covered by the pricing tiers, described by type of practice. The entry tier is for solo practitioners without team features, the middle tier for teams that share projects and prompts, and the top tier for firms and legal departments needing a full enterprise setup. Tasks are listed in a library of about twenty use cases, from clause extraction and contract review to due diligence, discovery summaries, litigation and matter summaries, horizon scanning, invoice summaries, redlining and NDA generation. Reach across countries shows indirectly, through English, German and Spanish sites, German and Austrian legal database links, and added terms for Austrian and Swiss professional secrecy holders. Missing: nothing names a practice area or firm type the product is not for, nothing covers government use, and jurisdictions are left to be inferred rather than stated.
Neota serves corporate legal departments, law firms (including services for their clients), and enterprise compliance, risk, procurement, HR, insurance and government teams. Case studies cover offshore funds, commercial contracting, matter onboarding and AML, freedom of information in the public sector and FCPA compliance. Customers are in North America, Europe and Asia Pacific. Nothing says what the platform does not support or where the AI tasks are unsuitable.
The 12 legal signals, side by side
Recorded rather than graded. These are the questions a practitioner has to answer before a tool touches a client matter, and the answers are taken from public material only.
Client Data in Training
Can material a lawyer puts into this product be used to train a model?
BRYTER's master agreement bans training on customer content. Its AI Terms section has a heading that says Customer Input will not be used for AI Training. The clause says BRYTER does not use Customer Content to train machine learning models, and that using the AI Service gives BRYTER no right or license to train on Customer Content. That second part closes off training by license as well as in practice. A separate clause licenses BRYTER to use Customer Data only to provide the software, which is therefore not a training license.
The ban is repeated on the pricing table for every tier and on the security page, which says customer data, prompts and outputs are never used to train AI models. The agreement is published, versioned and matches the marketing. The only exception: BRYTER may use data from the customer's use of the AI Service where the customer volunteers feedback or gives permission. Two further points. The customer may not use the AI Service to build competing foundation models.
And the addendum discloses that the chosen model provider screens inputs and outputs automatically, with human review of flagged content case by case. That is not training, but a third party may read customer content.
A public FAQ says customer data is not used to train models, with no matching term in a published agreement. The reason given is structural: Neota does not send customer data to a Neota model; customers connect their own model under their own provider agreement, inside their own AWS Bedrock or Azure boundary. No customer agreement is published, and the Terms of Use do not address training.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
BRYTER states retention periods category by category. A table in the data processing addendum gives purpose and duration for each type of data. Contact and account data and login credentials are kept until termination. Network and device identifiers, log files and audit trails are kept up to ninety days where applicable, otherwise until termination. Customer content, defined to include uploaded documents, free text, extracted fields and AI output containing personal data, is kept until termination and then deleted.
On termination BRYTER deletes, or returns and deletes, all personal data and processing results at the customer's choice, with backups allowed for thirty days under the addendum. So a buyer can trace a prompt from submission to deletion. Two frank disclosures in the same table qualify this. Data is also used for quality assurance, error analysis and monitoring of AI processing, described as LLM observability. And the provider of the chosen language model screens inputs and outputs automatically against its usage terms, reviewing flagged content case by case. No shorter retention window at the model layer is published.
The customer controls retention. The Privacy Policy says application data is encrypted, and when a session ends it is either deleted automatically or kept for a period the site owner directs. Zero retention is not stated as available for AI activity, because the audit trail logs every prompt and response, and how long that log is kept is not stated.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
A permission model of BRYTER's own is set out in its technical and organizational measures. Customer data is kept strictly separate, in separate systems, from BRYTER's own data, and customer data, test data, contact data and employee data are separated from each other. Rights management is central and split between system and application access. Users cannot change their own rights, and rights changes need approval. A role concept with phased access rights is based on identity management.
The agreement supports this. Each Authorized User needs an individual login, account sharing is banned, reassignment is restricted, users may only act within their assigned roles, and BRYTER has an audit right. Projects let users organize chats and documents by case. BRYTER integrates with iManage and SharePoint, but nothing says it enforces those systems' access rules per user at query time. Keeping the two aligned is the firm administrator's job. No conflicts or ethical wall feature is named.
The product keeps its own permission model, described at two levels. Building access is managed in Workbench with role based permissions and optional single sign on. Access to each application is defined by the customer down to who can start a workflow, see a matter or approve. It does not inherit permissions from the document system.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
Notice of compelled disclosure is a BRYTER commitment in two forms, with no transparency report. The master agreement allows disclosure under a valid order of a court, government or regulator, or as law requires. The receiving party must promptly tell the other party and let it object or seek a protective order, as far as the law allows. That is notice plus a chance to resist. The data processing addendum adds a second commitment for physical seizure.
If personal data in BRYTER's control is subject to search and seizure, attachment, confiscation in insolvency or similar measures, BRYTER must tell the customer without undue delay. It must also tell all parties that the data belongs solely to the customer, who is the responsible party. A third clause says that if law requires processing outside the customer's instructions, BRYTER must tell the customer first, unless the law forbids it on important public interest grounds. No transparency report, schedule or aggregate figure was found.
The Privacy Policy says Neota may disclose personal information in response to a subpoena, court order or other governmental request, with no commitment or reservation on notifying the customer. No customer agreement is published.
Primary Law Corpus Provenance
Where does the law in this product come from, and does the vendor have the right to use it?
Each content source is named with its rights basis, but BRYTER builds no legal corpus of its own. Instead it connects to named outside sources, with licensing stated on the pricing table. The Otto Schmidt legal database is reached by direct API on the top tier and needs an existing Otto Schmidt subscription. Haufe Lexware needs an added Haufe subscription on every tier. Handelsregister company data via Company.info and DeepL translation are paid extras, and DeepL also appears in the subprocessor list with its processing locations.
Separate terms are published for the Otto Schmidt Legal Data Hub. So the rights basis is that the customer brings its own license, a clear answer even though it shifts the duty rather than taking it on. Two limits. The web search feature runs through a named third party search processor, with no stated basis for what it retrieves. And nothing says what the language models were trained on, and the agreement reserves the right to change models at any time.
The applications run on the customer's own rules, documents and data with the customer's chosen model; no outside legal corpus is described.
Good Law Verification
Does the product tell you when the authority it just cited has been overruled?
BRYTER is a contract and workflow product first. Its use case library covers clause extraction, contract review and drafting, due diligence, document comparison, redlining, NDA generation, timelines and obligation tables, with legal research as one item of about twenty. Research runs through a licensed publisher's database the customer subscribes to separately, and any currency signals there are the publisher's, not BRYTER's. Nothing describes flagging overruled, superseded or negatively treated authority, and no citator is claimed.
Summaries keep citations attached, which shows where text came from, not whether the law is still good. The agreement says accuracy may be lower when access to legal sources is not enabled, which ties currency to a setting without describing any check.
No check of cited authority for subsequent history is described.
Refusal and Uncertainty Behavior
What does the product do when the answer is not in the corpus?
When the system cannot give a reliable answer, nothing says what it does, though BRYTER comes close. Hallucination control is a named feature on the security page but is never described: no abstention, confidence signal, threshold or handling of unclear input. A feature title is not a description. What the agreement does publish is when accuracy drops. The customer accepts that Output may be less accurate and reliable if web search or access to legal sources is not enabled, that the service is then more likely to be wrong, and that BRYTER recommends enabling them.
That is honest, but it describes a setting, not the model flagging doubt as it answers. The rest is disclaimer. BRYTER does not guarantee Output is accurate, complete, original, fair, unbiased or free of errors, and the customer must verify all Output before use. In practice, an Extract user running a large due diligence set is told to check everything, but not which extractions the system was least sure of.
Neota describes what happens when the AI is uncertain. Each AI task captures a confidence score as a workflow variable, and a decision gateway can branch on it against a threshold the customer sets. Low confidence results or failed checks can be routed to a named person, held, or returned to the requester for more information. The path is configured for each application rather than fixed.
Fabricated Citation Record
Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?
Searched on 12 September 2026, on the company name and the AI suite name, across published trackers and coverage of decisions on fabricated citations produced with AI. That included coverage of the Damien Charlotin database, reported as listing more than 1,353 affected filings worldwide, and reporting on 2025 and 2026 sanctions decisions in US federal circuits and state appellate courts. None located. This check covers fabricated citations only, so it is not a litigation history and other proceedings would not appear here.
The risk arises where a product generates legal authority for filing. BRYTER centers on contract review, extraction, drafting from the customer's own templates and workflow automation. Research against a licensed publisher database is available on the top tier, so the risk is real but narrower than for a research platform.
The AI Hallucination Cases database maintained by Damien Charlotin, and trade press reporting, record no court finding on fabricated or hallucinated content in output from Neota Logic.
Bar Guidance Alignment
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
On a dedicated page, BRYTER says BEAMON AI meets the data protection and professional law rules for Austrian lawyers, and names them: section 9 of the Rechtsanwaltsordnung, section 40 paragraph 3 of the RL-BA 2015, and the GDPR with the Austrian Data Protection Act 2018. It explains how, citing unlimited confidentiality, a processing agreement under Article 28 GDPR, EU data processing and an express ban on training AI models with customer data.
It says it has fully completed and signed, at company level, the Austrian Bar Association's checklist for AI providers, and that added provisions for Austrian professional secrecy holders apply. Those provisions are published as a standing document in its legal terms. Germany is covered more briefly: the security page states compliance with the BRAO and section 203 of the Criminal Code, naming the rules without mapping the product to them, and the agreement offers a discount tied to bar association membership. BRYTER also publishes notes for Swiss and German professional secrecy holders.
Professional responsibility is referred to in general terms. Neota's home page says human oversight ensures decisions meet legal and professional standards, and its FAQ lists regulators, clients, courts and insurers as parties who may ask for the audit trail. No ethics opinion, court AI order or rule, or judicial or bar guidance is named, including ABA Formal Opinion 512.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
Time savings are the claim, the product sits inside the fee relationship, and nothing addresses billing or disclosure. The claims are clear: save time and reduce costs, save hundreds of hours reviewing contracts, free legal teams from repetitive tasks, and speed up work that took thousands of hours of manual review. Buyers include global law firms that bill those hours to clients, so this is BRYTER's own selling point.
Nothing addresses what happens to the bill. There is no per matter record of AI work, nothing marks output as machine produced for fee purposes, and no fee or disclosure guidance appears, even in a detailed agreement of sixteen sections with its own AI terms. Two features touch invoices but are not records of AI work: a billing guidelines agent that reviews and corrects invoices and matter descriptions, and an invoice summary use case. One usage rule points this way: a contractual bar on misleading anyone that Output was written by a person.
A usable record of work done with AI exists, with no published fee guidance. Law firms are among the buyers and build services for their clients on the platform. The audit trail records each AI call by session and matter with its token count and cost. Nothing addresses how firms should bill or disclose work done with AI. The Harneys case study describes that firm moving to fixed fees on its own account.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
All three items a firm would forward are available without an agreement or request. Schedule 1 of the data processing addendum lists each subprocessor with its service, company location and server location: cloud hosting in Frankfurt, a monitoring tool, a file conversion processor, a translation provider and a web search processor. The same table names the model providers. Microsoft Azure hosts the default language model in BEAMON, and Anthropic Ireland is an optional model used only when a customer administrator enables it.
A firm asked which third party sees its client content can answer by name, role and location, and say which are on by default and which need its own action. The addendum itself is the client facing document, drafted to be handed to a client. Subprocessor changes need prior written notice with a twenty business day objection right. Two further points. The addendum bars subprocessors from passing on professional secrets. And a completed bar association checklist for AI providers exists in one country as a forwardable document.
Disclosure material exists behind a request. The Security page offers a security pack through the contact form, and no subprocessor list is published. The model providers are named on the Integrations page, but they run under each customer's own account, so a firm would disclose its own provider rather than one chosen by Neota.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
Disclosure of AI involvement in legal work has no BRYTER feature behind it, though two things come close. The security page says access is limited by role and need, actions are logged and changes traceable, and lawyers stay responsible for decisions, with audit trails supporting internal governance and external requirements. But nothing says the trail shows which passages a model wrote, separates machine drafted from human edited text, survives export, or could be put before a tribunal.
It reads as a security log, not a disclosure record. The second is a usage rule: the customer must not mislead anyone that Output was written by a person. That is a duty of candor on the customer, not a feature that makes candor possible. There is no disclosure template, certification, model identifier on output, or export built for filing. This matters because BRYTER's agents handle court documents and claims, so output does get near a tribunal. A firm later asked which parts of a filing a model drafted would have nothing published to work from.
An exportable record for each session covers the AI's part in the work. The AI Solutions page shows each AI call recorded against its session and matter with the prompt version, the clause and citation captured, the model used, tokens, confidence and the approving person. The record is hashed as immutable, and the page says the whole trail is exportable.
The questions both sides leave open
Derived from the records above rather than written, so it cannot favor either vendor. Take these into both conversations and ask each side the same question.
- Good Law Verification
Product updates
The latest logged changes from each vendor, newest first, each linked to its source.
BRYTER
-
Product / capability
BRYTER released BEAMON 2.0, which adds an agentic mode in which the assistant plans a multistep task, pulls information from several sources and works through it to completion, such as drafting a long pleading to a detailed specification or researching across many documents.
Source
In summary
BRYTER
BRYTER, from BRYTER GmbH of Frankfurt am Main, founded in 2018, sells two products. BRYTER Workflows is a no code platform on which legal and compliance teams build intake, approvals, document automation and client portals. BEAMON AI is a suite of an assistant, bulk document extraction and a Microsoft Word integration for drafting and review. According to the AI Legal Index, its agreement bars training on customer content, its data processing addendum carries a professional secrecy section, and BEAMON AI is priced from €99 per user a month. It names Linklaters, Ashurst and KPMG among customers.
Neota Logic
Neota Logic, from Neota Logic Inc. of New York with group companies in London and Melbourne, is a no code platform on which legal, compliance and risk teams build intake and triage, contract generation, regulatory assessments and client facing advice tools. It is now positioned as a governance layer for AI. Its AI Task Node calls a model the customer has approved, through the customer's own provider account, to extract, classify, summarize or draft inside a deterministic workflow. According to the AI Legal Index, a confidence score can route any AI step to a reviewer, and each call is logged with the model that ran it. It states ISO 27001 and names Harneys and Fujitsu among customers.
Questions buyers ask
BRYTER vs Neota Logic: which is better for legal workflow automation?
They are close, and the difference is where each puts its weight. BRYTER publishes more in its contract, including a professional secrecy clause, and prices its AI suite openly. Neota Logic gives the customer more control over which model runs and when a person must review its output. From the AI Legal Index, based on each vendor's own published materials as of September 26, 2026. No vendor pays for placement.
Does BRYTER train AI on customer data?
No, by contract. The AI terms in BRYTER's master agreement state that it does not use customer content to train machine learning models and that use of the AI service grants it no license to do so. The same statement appears on its pricing table and security page. Neota Logic's FAQ says customers connect their own model under their own provider agreement, so no Neota model is trained on their data. From the AI Legal Index, based on each vendor's own published materials as of September 26, 2026. No vendor pays for placement.
Which AI models do BRYTER and Neota Logic use?
BRYTER's subprocessor schedule names Microsoft Azure as host of its default model, processed in the EU, and Anthropic as an optional provider an administrator must enable. Neota Logic names OpenAI, Anthropic Claude, Google Gemini, Azure AI Foundry, Amazon Bedrock and self hosted models, run through the customer's own account, with the model recorded on every call. From the AI Legal Index, based on each vendor's own published materials as of September 26, 2026. No vendor pays for placement.
How much do BRYTER and Neota Logic cost?
BEAMON AI, BRYTER's AI suite, costs €99, €149 or €199 per user a month on annual plans and €118, €176 or €234 monthly; BRYTER Workflows is quoted. Neota Logic prices per solution rather than per user, on a tiered subscription whose cost per solution falls as more are built, plus fixed fee professional services. From the AI Legal Index, based on each vendor's own published materials as of September 26, 2026. No vendor pays for placement.
Neither vendor paid for inclusion, placement or a grade, and neither reviewed this page before it published. Everything on it comes from public material on the dates shown. How the index grades.
Three points to weigh. BRYTER's own documents disagree on data location. One schedule places processing in Frankfurt and the EU, another in United States data centers. Its security page says data never leaves the EU, while its subprocessor list includes optional US providers. BRYTER's support commitments exclude its AI from guaranteed response times. Neota Logic publishes no customer agreement, so its liability position and training commitment sit outside a readable contract. Neither vendor reviewed this page.