Annlex vs Descrybe: how they compare in 2026

The metered unit is the model. Every paid plan, the credit pack and the free allowance count AI queries, which the pricing page defines as each Annlex Chat question, each AI case analysis and each synthesis request. Boolean, semantic and hybrid search are free and unlimited. Customers pay for Annlex Chat's retrieval and synthesis over 10.5 million opinions and for the Opinion Lens tools (Key Passages, Summary, Case Brief, Ask AI). Even the free semantic search runs on ModernBERT embeddings fine tuned for legal text by the Free Law Project. Without the models, what remains is a free search box over a public collection Annlex does not own.

The content itself is model output, which is as central as AI can be. The product was built by AI summarizing judicial opinions, now 10,969,112 opinions across 492 courts and tribunals per the July 2026 coverage page, and that summarization is how the database was built, not a feature on top of it. On top sit DescrybeLM, a purpose built legal reasoning engine; Cytator, an AI citator giving treatment analysis by issue; and an engine layer offering retrieval and verification tools to outside assistants. Without the models there would be no summaries, no citator and no product, only public domain opinions the company never had rights to sell.

Grounding is real and documented, and Annlex is candid about its limits. The FAQ (version 1.2, March 2026) describes the method. Each question is classified, matched to one of ten research patterns and sent through a pipeline that retrieves opinions before the model writes. Every claim must cite a case from the database, and every citation opens the full opinion, which Annlex holds on its own servers from the CourtListener bulk export. Answers carry a Verify link into Opinion Lens, where a characterization can be checked against the opinion passage by passage. Annlex names the failure grounding leaves: a real case cited correctly and described wrongly, a dissent quoted as the majority, a rule applied in the wrong jurisdiction. The terms list misread precedent, missed or inappropriate authority, outdated law and missed controlling authority as expected failures. No measured accuracy is published, with no test set, citation resolution rate or characterization error rate, and returned authority is not checked for subsequent history. The FAQ's opening answer says the design eliminates the conditions that cause hallucination, while the same page and the terms describe a constraint that shifts the failure mode instead.

Grounding is well structured, without a published measurement. DescrybeLM answers carry a table of authorities linked to primary law, the reasoning is shown at each step, and quote verification and citation resolution are built in as tools. The vendor also discloses against its own interest: the terms say Descrybe does not review or validate outputs for accuracy, truthfulness, relevance or reliability, or to confirm they are not misleading, and earlier material warned users to check summaries against the original opinion. No accuracy figure of any kind is published: no summary fidelity rate, citation resolution accuracy, hallucination rate or test set. A claim in third party material that DescrybeLM beats named frontier models on bar exam benchmarks was not found in vendor material with any method or results, and is not relied on.

The product is built around a person checking the machine, and the review surfaces are real. Opinion Lens keeps the opinion on screen at all times, and the generative tools sit beside the text and point to the paragraph an answer rests on. Key Passages anchors the question presented, rule, holding and limitation to paragraph numbers, and Ask AI declines when the opinion does not contain the answer. Annlex Chat links each citation to its opinion and offers a Verify route into Opinion Lens. Annlex's FAQ states that attorneys remain responsible for independently verifying all authority before relying on it, that the tool suits preliminary research rather than replacing verification, and that it should not be anyone's sole source. Nothing acts on a matter; output is research text and downloadable summaries and briefs. The control structure stops short. No threshold at which Annlex Chat declines is described, the FAQ's promise to show confidence levels where applicable is not explained, and nothing addresses what happens after an answer is found wrong.

Oversight is described as product behavior rather than as policy, which is the useful form. DescrybeLM is documented to clarify jurisdiction, posture and key facts with the user before proceeding, which puts a human checkpoint at the start of the task rather than only at the end, and it is documented to tell the user when it needs more information and to flag uncertainty rather than filling gaps with guesses. Reasoning is visible at each step and work product is exported deliberately to Word rather than acted on. In the engine configuration the user can see which Descrybe tools an external assistant called and read the data each returned, which is oversight of a tool chain and not merely of an answer. None of it is quantified or bounded: no statement of when the system proceeds without clarification, no confidence threshold, and no description of what the uncertainty flag actually triggers.

No production evidence is published: no named customer, firm, law school, library or user count, no case study, testimonial or usage figure, and no outcome measure appears on the Annlex Chat page, pricing, the FAQ, the terms or the Opinion Lens page. The sitemap lists no customer, news or blog page. The product is live with published self serve pricing, and the founder's public profile records a pitch at the Dig South Tech Summit in Charleston, which is launch activity rather than evidence of deployment.

Institutional adoption claims are specific, but none was found in vendor material. Third party research says Descrybe was added to the NSLT curriculum in place of Casetext and approved as a NELLCO electronic resource reaching nearly 150 law libraries. Those would be strong evidence if the vendor published them, since library consortiums and curricula evaluate research tools carefully, but as found they are third party claims. Dated and checkable: coverage by a named legal technology journalist across four product stages from 2024 to 2026, an ABA Woman of Legal Tech award for cofounder Kara Peterson in 2024, and an Anthem Award for Responsible Technology. No customer, deployment, usage figure or outcome measure published by the vendor was found, and the coverage milestones describe the vendor's own build, not customer results. The source basis is recorded as third party estimated.

Annlex writes real commitments on training and on the model provider into its terms; separation between users is not addressed, and the account of storage is inconsistent. Section 10 of the terms (last modified date shown as 27 February 2026) says saved conversations are not used to train any AI model, Annlex's or anyone else's. Annlex Chat history saves by default to Annlex's own servers for registered accounts, is kept up to one year and then deleted, and can be switched off or deleted immediately. Queries go to Anthropic's API, which does not train on API data and may keep query content up to 30 days, longer where a safety system is triggered. Zero Data Retention with Anthropic is stated to be not yet active. The FAQ ties attorney client privilege to that future arrangement, as positioning rather than a stated handling position. Nothing addresses separation between users or any team or matter structure. The terms' third party section says chat text is stored in no database, and the pricing FAQ says queries are not stored; both conflict with the history paragraph. No data processing addendum or separate customer agreement exists for a firm to sign.

Published commitments cover what firms ask first, but not privilege itself. The security page, updated 30 July 2026 and read 11 September 2026, says customer prompts, files and history are not used to train Descrybe owned models; OpenAI's API is the main model provider, opted out of training, with abuse logs kept up to 30 days; uploaded briefs are not kept once text is extracted; users can delete research history at any time; and the product uses individual accounts with no shared or matter workspaces, so separation is per user and matter walls do not exist. The binding terms of service, described as the legal equivalent of a signed contract, say nothing on these points, so the commitments are policy, not contract. Privilege and work product are not addressed anywhere, including the privacy policy (updated 29 September 2025) and the security page, though the product now takes in users' briefs and pleadings. Where OpenAI processes requests is not stated.

Annlex takes a real position on the advice line and repeats it on the product page, without full treatment. The terms open with a section saying the service is not your lawyer, calling it an experimental research tool that does not give legal advice and directing users to a licensed attorney. The FAQ adds that payment does not create an attorney client relationship and lists prohibited uses, including reliance as legal advice, sole source research and FCRA purposes. The Annlex Chat page, which names pro se litigants first among its audiences, says it is a research tool only, not legal advice, and to consult a licensed attorney, so the notice sits at the point of use, facing consumers. Competence is covered in outline: attorneys are told they remain responsible for verifying all authority before relying on it. No jurisdiction limits are named beyond the collection being American case law, bar guidance is left to each practitioner's jurisdiction, and nothing covers supervising the paralegals Annlex lists among its users.

Not addressed, on a product where it most obviously should be. The vendor says the product brings legal reasoning to anyone who needs it, lawyers and nonlawyers alike, and its founding mission is access to justice for people without legal training. A guided reasoning engine that takes a member of the public's question and returns a cited analysis raises the unauthorized practice question directly, and no position was found: no statement that output is not legal advice, no guidance on when to consult a lawyer, and no engagement with any bar rule. The closest is the terms disclaiming review or validation of outputs, which allocates risk without addressing the professional question.

Nothing is published: no AI policy, evaluation or release testing, bias or accuracy monitoring, model card, named owner or outside standard. The nearest material is a list of limitations in the terms and FAQ and a design statement on the Opinion Lens page that credibility rests on falsifiability, which describes a product choice rather than a governance mechanism. The gap matters here because the ten research patterns that route every question decide which opinions the model ever sees, and nothing describes how those patterns were built, tested or changed.

Nothing is published on governance: no AI policy, evaluation method, bias or fairness assessment, accuracy monitoring, drift statement, model card, governance body or external standard. The gap matters here because every summary is machine generated, so any systematic bias in summarization flows into every search result and citator treatment, unseen by a user who reads the summary rather than the opinion. The vendor has already rebuilt all its summaries once on newer models, showing that quality depends on the model, and published no evaluation of either version. On 11 September 2026 the privacy policy and the security page, updated 30 July 2026, were also read; the security page says models may be updated for quality, safety, performance or reliability without notice unless the change is material, which concerns the supply chain, not governance.

Retention and deletion are specific enough to hold Annlex to; the rest is thin. Saved conversations are kept up to one year and then deleted automatically, a user toggle stops saving, saved conversations can be deleted immediately and irreversibly, and no history is saved for users without an account. Google Analytics data is deleted after 26 months with IP anonymization, and server access logs are said to hold request metadata only. Named third parties are listed with their data terms: Anthropic (no training on API data, retention up to 30 days, longer on a safety trigger), Stripe for payments and Google Analytics. Access, correction and deletion requests go to a published privacy address. There are four gaps. The server infrastructure provider is referred to but not named, so the subprocessor list is incomplete. No access control practice is described. Security is stated only as reasonable measures including HTTPS, with nothing on encryption at rest. No incident or breach practice is published. The terms' third party section, which says chat text is kept in no database, also conflicts with the default history described earlier in the same document.

Most of the ground is covered, specifically enough to hold the vendor to. The security page, updated 30 July 2026 and read 11 September 2026, says what is kept (prompts, research history, outputs, metadata and logs) and what is not (uploaded files are discarded once text is extracted). Users can delete history at any time and accounts on request; no zero retention mode is offered; production access is limited to authorized technical staff on a need to know basis; and data is encrypted in transit and at rest on DigitalOcean managed services in the New York region. It names subprocessors with links to their attestations: OpenAI's API, DigitalOcean, Clerk and Stripe, with Google Analytics in the privacy policy. OpenAI API training is opted out. Two gaps remain: no incident or breach practice is published, and no retention period is given for logs and metadata outside the user's deletion control.

Liability is handled through a standard clause that disclaims all of it. The terms provide the service as is with every warranty disclaimed, and exclude liability for any direct, indirect, incidental, special or consequential damages from use, with no cap figure because nothing is retained to cap. Disputes go to binding AAA arbitration under the Consumer Arbitration Rules seated in Charleston, South Carolina, with class action and jury waivers, under South Carolina law. The recourse that exists concerns availability, not output: prorated credit or refund for a substantial outage, full refunds for billing errors, and 30 days' notice with prorated refunds if the service is discontinued. The terms list the ways AI output can be wrong and put reliance at the user's risk. No indemnity, accuracy undertaking or correction route for wrong output is published.

A published liability position exists and runs almost entirely one way. The terms of service make the user indemnify and hold Descrybe harmless for liability, loss, claims and expenses, including reasonable attorneys' fees, related to use of the service or its outputs, and limit any Descrybe liability to what the user paid in the twelve months before the claim. At published prices, that caps vendor exposure at $300 to $600 per user. The terms also say Descrybe does not review or validate outputs for accuracy. A specific, readable cap is real disclosure a buyer can price. No warranty, accuracy undertaking, service level or correction route was found.

No integration into practice systems is published: no document management, practice management, Word or Outlook add in, API or connector. Summaries and case briefs download as formatted documents, which is export rather than integration. The absence is a stated choice: the FAQ argues that tools reaching Claude through MCP connectors cannot qualify for Anthropic's Zero Data Retention while a direct API setup can, so Annlex has chosen not to be an assistant plugin.

Integration is the product, documented at tool level. Descrybe Legal Engine offers citation lookup, quote verification, treatment review and source retrieval to outside assistants over MCP, naming Claude, ChatGPT and Perplexity as supported hosts, plus an Open Connector for building more tools. A user guide describes the flow: the assistant calls one or more Descrybe tools, Descrybe runs the search, lookup, verification or retrieval and returns focused results, and the user can see which tools were called and what each returned. Work exports to Word. The integrations are to AI assistants, not legal practice systems: there is no connection to document management, practice management or a firm's matters.

Annlex is a hosted web application, and little else is stated. The case law collection sits on Annlex's own servers, and research queries are processed through Anthropic's Claude API. No hosting provider, region, tenancy model or processing location is named, and the terms refer only to an unnamed server infrastructure provider. There are no deployment options to compare.

Tenancy and storage region are stated; processing location is half stated. The security page, updated 30 July 2026 and read 11 September 2026, says customer data is stored on DigitalOcean servers in the New York City region, the product runs on provider managed multitenant infrastructure, it is built around individual user accounts, and no single tenant, private or shared workspace option exists. Research requests are processed by OpenAI's API, and nothing says in which region. Queries routed through an outside assistant such as Claude, ChatGPT or Perplexity add a leg whose location is also not described. No residency options are offered.

No independent security attestation is published. No SOC 2, ISO 27001 or other certification is claimed, there is no trust center, and the terms describe security only as reasonable measures including HTTPS encryption, with no guarantee of absolute security. Stripe's PCI DSS standards are cited for payment handling, which covers Stripe, not Annlex.

The vendor states plainly that it holds no attestation of its own. The security page, updated 30 July 2026 and read 11 September 2026, says Descrybe does not hold its own SOC 2 or ISO 27001 certification and has begun work toward SOC 2. It relies on providers' programs, linking to trust portals for DigitalOcean (SOC 2 Type II, SOC 3), OpenAI (SOC 2 Type 2, ISO 27001, 27017, 27018, 27701), Stripe (PCI Level 1, SOC 2 Type 2) and Clerk (SOC 2 Type 2). Data is encrypted in transit over TLS and at rest on DigitalOcean, and production access is need to know. A provider's report covers its infrastructure, not the application layer where Descrybe's code, access controls and data handling sit, which is what a firm's security questionnaire asks about. Descrybe is a self funded two founder company charging $25 to $50 a month, and a SOC 2 examination is a material cost against that model. The security page sits in the site footer.

The models are named; notice of changes is the gap. Annlex's FAQ states that Annlex presently runs Anthropic's Claude Sonnet as its leading model, calling the Claude Messages API directly, and that semantic search uses ModernBERT embeddings fine tuned for legal text by the Free Law Project. It describes Annlex's own contribution as an orchestration layer that directs leading models. The provider's data terms are stated alongside: no training on API data, retention up to 30 days, and Zero Data Retention not yet active. No model version is pinned, the word presently signals change without any commitment to tell customers when it happens, and where the provider processes requests is not stated.

The provider is named, and change notification is the gap. The security page, updated 30 July 2026 and read 11 September 2026, says Descrybe uses OpenAI API models for summaries and research outputs, that OpenAI's API is its main LLM provider, that the exact model may vary by feature and change over time for quality, safety, performance or reliability, and that users are not notified of each update unless it materially affects terms, privacy or data processing. This settles earlier conflicting signals: the terms disclose third party models without naming them, and a third party directory had called DescrybeLM the company's own model. The page also says OpenAI training is opted out and abuse logs are kept up to 30 days, linking to OpenAI's data controls. No specific model or version is named, there is no commitment to notify of model changes, and where OpenAI processes requests is not stated.

The full price list is public and the unit is defined. Starter is $25 a month for 200 AI queries and Power is $75 a month for 600, both with a seven day trial that requires a card; a one time pack of 100 queries costs $10 and never expires; three AI queries are free without an account, and Opinion Lens allows three questions per opinion before signup. Boolean, semantic and hybrid search are free and unlimited. The pricing page defines what counts as a query, and the terms publish trial conversion, the absence of rollover, 30 days notice of price increases and the refund rules for packs and subscriptions. Signup is self serve, so nothing is added for implementation, and only volume beyond the Power plan is by contact. One loose end: the refund terms refer to annual subscriptions, and no annual plan is published.

Exact prices are published for every tier: $25 per month for Descrybe Legal Engine and $50 per month for the full Descrybe Platform with the Engine included. The unit is a monthly subscription per user. The vendor says every feature is included at either tier, so the headline price does not buy a limited subset. It also explains its pricing: it is self funded, without outside investor timelines or enterprise sales costs, and prices for the work rather than the market. A buyer can work out total cost from the public site without contacting anyone.

Audience and use limits are stated precisely; the limits of the case law collection are not. Annlex names pro se litigants, law students, solo practitioners and small firms priced out of Westlaw and Lexis, paralegals and journalists. It lists appropriate uses (case discovery, preliminary research, academic work, journalism) and prohibited ones (legal advice, sole source research, FCRA purposes), and says plainly the tool suits preliminary research rather than replacing verification. The collection is described as 10.5 million full text opinions from 3,355 federal and state courts, 1658 to the present, with 71 million citation links and a taxonomy of 20 legal areas and more than 1,125 concepts, updated quarterly from CourtListener's bulk export. Annlex's own site does not list courts or date ranges by court, say what curation from CourtListener left out, or say that statutes, regulations and secondary sources are not included. The FAQ also describes opinions as retrieved live from CourtListener, while the Annlex Chat page and the quarterly cadence describe a stored copy.

Coverage is listed court by court, with limits stated. The coverage page (descrybe.com/coverage, data as of 18 July 2026) lists 492 courts and tribunals and 10,969,112 opinions: federal (137 courts, 3,221,272 opinions), state (201 courts, 7,324,869), bankruptcy (97 courts, 80,941), specialized and agency tribunals (47 tribunals, 340,850) and tribal courts, plus statutes, regulations, constitutions, session laws and court rules. Each court shows its opinion count and first and latest dates, so a researcher can check whether a recent decision should be present. Most active supreme and appellate courts run to the first days of July 2026, and gaps are shown, such as Connecticut Superior Court ending in December 2014 and the Alabama Court of Appeals in 1999. No refresh schedule is stated beyond the as of date, and a few rows, including the New Hampshire district court and the Minnesota and Arizona appellate courts, show latest dates in the future, which look like source metadata. The vendor says it serves lawyers and nonlawyers alike.