Mentat
Also known as: 멘타트, Mentat Legal, Mentat AI, NEXUS, Mentat Co., Ltd., 주식회사 멘타트
Early stage Korean legal agent that drafts litigation documents, complaints, answers, and preparatory briefs, from uploaded materials, with a full self drafting mode and a paragraph by paragraph co pilot.
Mentat is a Seoul based AI company, Mentat Co., Ltd., building agents for regulated industries in Korea. Its NEXUS platform turns what a company knows and what it must comply with into governed intelligence.
Corpus gathers company knowledge from shared drives, file servers, wikis and mailboxes behind department and role permissions enforced at query time, Vault holds each user's or each case's own files in an isolated collection, and Codex is a knowledge graph of every current Korean statute with its subordinate rules, official interpretations, precedents and sanction cases, so a single question returns the whole body of law that applies.
Prebuilt search, research and vault agents answer with sources, custom agents are composed from a marketplace of plugins and skills, and Workflow saves repeatable jobs such as credit reports and opinion letters as plan, execute and verify sequences that run on a schedule with a person approving each step and every output recording its grounding.
Harness, the orchestration engine underneath, splits work across role separated sub agents, calls statute databases, document generation, spreadsheets and the customer's in house APIs through schema defined tools with per tool permissions, checkpoints every run and keeps a tamper evident audit log of who decided what and why. Mentat AI is the litigation product built on the same stack.
A lawyer uploads case files in HWP, PDF, Word, image or audio form, up to a thousand pages, and either lets Full Self Drafting analyze the issues, design the outline and draft a complaint, answer or brief from a stated strategy, or works paragraph by paragraph in Co-pilot mode confirming each change, across twenty six document types with case law linked. Sixty lawyers ran 832 cases through a 218 day beta before the May 2026 launch. NEXUS deploys on premises, in a private cloud or hybrid inside network separated environments and is sold by demo. Mentat AI is a monthly subscription from ₩39,000 for Korean licensed attorneys, with a free first month on the Pro plan.
Vendor details
Canonical URL
https://mentat.co.kr
Category
Enterprise operations agent
Funding status
Seed (about 800 thousand dollars, 1.1 billion won), 2026
Company status
independent
Use cases & customers
Primary use cases
Target customers
Integrations
Drafts litigation documents from materials the lawyer uploads; system integrations, tool calling, and APIs are not documented.
In practice
A litigation team needs a first draft of a brief. The lawyer uploads the case files in HWP, PDF, Word, spreadsheet, image or audio form, and Full Self Drafting analyzes the issues, designs the outline and drafts the document, or the lawyer works in Co-pilot mode and confirms every addition or deletion sentence by sentence.
A compliance team needs every rule that applies to a new product. NEXUS answers from Codex, a knowledge graph of all current Korean statutes with their subordinate rules, interpretations, precedents and sanctions, and from the company's own documents in Corpus, and every answer records the sources it drew on. Documents outside a user's rights never enter retrieval.
A credit team runs the same report every week. NEXUS Workflow saves the steps and judgment criteria and repeats the run on a schedule, Harness halts high risk judgments until a person approves, and the audit trail records who decided what, when and why.
Sources & related URLs
Related / legacy domains
Agentic Index coverage score
9.0 / 14 capabilities · 64%
| Integrations & Tool Calling | Full |
|---|---|
|
Agents use a schema defined tool interface to read statute and case law databases, write documents, run spreadsheet calculations and call the customer's in house APIs, with approval required. Each tool carries its own permissions and call limits. NEXUS Corpus ingests shared drives, file servers, wikis and mailboxes. SourceMentat, mentat.co.kr/products/harness and mentat.co.krread 2026-09-06 |
|
| Workflow Orchestration | Full |
|
Harness breaks a task into a plan and dispatches role separated sub agents for planning, generation, evaluation and synthesis, each with its own prompts, tools and token budget. It verifies against criteria and re plans shortfalls until the work converges. NEXUS Workflow saves those step sequences and judgment criteria as repeatable automations, and Mentat AI's Full Self Drafting runs issue analysis, outline design and drafting end to end. Customers get a workflow builder for these multi step flows. SourceMentat, mentat.co.kr/products/harness and mentat.co.krread 2026-09-06 |
|
| Knowledge Grounding & RAG | Full |
|
NEXUS keeps three stores. Corpus is a permission aware store of the company's knowledge built from shared drives, file servers, wikis and mailboxes. Vault holds isolated collections for each user or case. Codex is a knowledge graph of all current Korean statutes with subordinate rules, interpretations, precedents and sanctions. Every answer records the sources it drew on, and Mentat says more than ten million documents are in operation. SourceMentat, mentat.co.kr and mentat.co.kr/products/mentat-airead 2026-09-06 |
|
| Human Oversight & Guardrails | Full |
|
Harness places stop and approval points at explicit plan, execute and verify gates. It halts high risk judgments until a person approves, escalates to a person when the verification loop exceeds its iteration limit, and restricts each step to the tool permissions it needs. In Mentat AI's Co-pilot mode the lawyer confirms every AI addition or deletion sentence by sentence. These gates are built into the product. SourceMentat, mentat.co.kr/products/harness and mentat.co.kr/products/mentat-airead 2026-09-06 |
|
| Security, Identity & Governance | Full |
|
Department and role permissions are enforced at query time, so documents outside a user's rights never enter retrieval or generation, and the setup is compatible with Chinese walls. Personal Vaults are indexed only from consented folders and readable by no one else. NEXUS integrates with the customer's existing authentication and access control, encrypts data in transit and at rest, and keeps customer data in the customer's environment without training on it. There is no published SOC 2, ISO or other attestation. SourceMentat, mentat.co.kr/securityread 2026-09-06 |
|
| Observability & Auditability | Full |
|
Tamper evident audit logs cover every data access and model call. They replay which query reached which documents under which permissions and can be extracted for a regulator. Harness keeps checkpointed execution traces recording who decided what, when and why at each plan, tool call, verification and approval step, and every output carries source citations. SourceMentat, mentat.co.kr/security and mentat.co.kr/products/harnessread 2026-09-06 |
|
| Memory & State Persistence | Partial |
|
Run state survives through checkpoints. Harness saves the full execution state so a run can stop and resume without losing it, and Co-pilot drafting holds a conversation about a case based on its documents. That is state within a run or session. There is no memory layer with a stated scope and lifetime across runs, and Vault and Corpus serve as knowledge stores, not memory. SourceMentat, mentat.co.kr/products/harnessread 2026-09-06 |
|
| Deployment & Data Residency | Full |
|
NEXUS has three deployment options. It can run fully isolated on premises in the customer's data center, inside network separated environments built to Korean financial sector rules. It can also run in a private cloud on a dedicated virtual network, or as a hybrid with sensitive data and models kept inside. Mentat AI is offered as a hosted SaaS. SourceMentat, mentat.co.kr/securityread 2026-09-06 |
|
| Prebuilt Agents / Templates / Packs | Full |
|
Mentat AI ships twenty six litigation document types a lawyer selects among, and NEXUS ships three prebuilt agents (unified search, research, vault) alongside a marketplace of plugins and skills for composing custom agents. Each document type is a complete template on its own, while the marketplace holds plugins and skills, not whole agents. SourceMentat, mentat.co.kr/products/mentat-ai and mentat.co.krread 2026-09-06 |
|
| Triggers & Channel Coverage | Partial |
|
NEXUS Workflow saves a workflow and repeats it on a schedule, such as a weekly credit report run. Users start work from the NEXUS question window or the Mentat AI web app. There is no event trigger, webhook or inbound channel beyond the web application. SourceMentat, mentat.co.krread 2026-09-06 |
|
| Model Flexibility & Routing | Not documented |
|
Mentat says the language model layer never trains on customer data, and the hybrid deployment keeps models inside the customer's environment. No model provider is named, there is no sub processor or model disclosure, and customers and admins cannot select a model. SourceMentat, mentat.co.kr/security and mentat.co.kr/products/harnessread 2026-09-06 |
|
| APIs / SDKs / MCP Extensibility | Not documented |
|
Outside systems have no way in. Harness calls outward to the customer's systems and the marketplace takes plugins and skills in, but there is no API, SDK or MCP server through which an outside system calls the Mentat platform, and no developer subdomain. SourceMentat, mentat.co.kr/products/harnessread 2026-09-06 |
|
| Testing, Debugging & Optimization | Not documented |
|
Harness runs an evaluator agent that scores each generated output against criteria and sends it back for rewriting until it passes. That loop runs on every output at run time. It is not a harness that tests a change to a workflow or agent. There are no test cases, evaluation dashboard or release gate for the workflows customers build. SourceMentat, mentat.co.kr/products/harnessread 2026-09-06 |
|
| Browser / Computer-use | Not documented |
|
Mentat AI and NEXUS run as web applications that take uploaded HWP, PDF, DOCX, spreadsheet, image and audio files and return drafts and answers inside the app. Agents reach outside systems through schema defined tools, and there is no browser, desktop or computer control. SourceMentat, mentat.co.kr/products/mentat-airead 2026-09-06 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
From ₩39,000/month
credits
Cost watchouts
As a usage based model, cost scales with drafting volume; published rates are not yet available.
Variable cost rationale
A usage based model means cost scales with drafting volume; published rates are not yet available.
Sales call required
No, self serve available
Free / trial
Pro carries a free first month as a launch offer; no free tier
Lowest paid plan
Standard, ₩39,000 per month excluding VAT, 4,000 credits and 2 vaults
Key ambiguities
The service is newly launching, so a usage based model is announced but exact prices are not yet public.
Related vendors
- 11th Estate — Agentic platform whose AI engine scans markets, matches an…
- Adopt AI — AI CPA firm whose agents run reconciliations, close, AP and AR, and…
- Aera Technology — Decision intelligence platform where an always on agent executes…
- Akro AI — On premise operational intelligence that automates document heavy…
- alfred_ — AI executive assistant that triages the inbox overnight and scores…
- Alloy.ai — Commerce intelligence system for consumer brands that unifies four…
Alternatives to Mentat
The closest documented capability profiles to Mentat among enterprise operations agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Variance10.0 / 14Adds documented APIs, SDKs & MCP Extensibility and Browser & Computer Use
- Clio10.5 / 14Adds documented APIs, SDKs & MCP Extensibility
- DraftWise8.5 / 14Adds documented APIs, SDKs & MCP Extensibility
- Legora10.5 / 14Adds documented Model Flexibility & Routing and APIs, SDKs & MCP Extensibility
- Workday10.5 / 14Adds documented APIs, SDKs & MCP Extensibility and Testing, Debugging & Optimization
- CoCounsel10.0 / 14Adds documented Model Flexibility & Routing and APIs, SDKs & MCP Extensibility
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded