Back to vendors
U

Undermind

Also known as: Undermind AI, Undermind Deep Search

Visit site
Entry priceIndustry Pro $75 per month, Academic Pro $20 per monthFull pricing detail

Undermind investigates scientific questions across published papers and a researcher's saved work. Its search agent plans queries, follows citations, and brings relevant sources into a persistent workspace. It can retrieve open access papers, read figures and tables, and save reports that link conclusions to the supporting text. Users can monitor a saved topic for new publications or connect the same workspace and research tools to compatible outside assistants.

Undermind gives researchers a workspace for literature discovery and synthesis. A user explains the evidence they need, and a deep search plans multiple queries and follows references, citations, and authors. The system repeats those steps as it discovers relevant material and returns ranked papers. Researchers can organize the results in a workspace, keep the source documents, and ask follow up questions about specific claims, tables, or figures. A report can link its conclusions to the papers used to support them.

The workspace retrieves open access PDFs and accepts papers a researcher provides. A team can reuse material from its library for later questions and compare what several papers report without starting a new search each time. Publication monitoring lets researchers receive updates when relevant work appears. Enterprise deployments can use internal documents alongside published literature and send updates based on a saved topic. A compatible outside assistant such as Codex, Cursor or Perplexity can connect to the same workspaces under the user's identity and call the research tools to find, read, and organize evidence.

Undermind offers Free, Pro, Team, and Enterprise options. Academic Pro costs $20 per month with monthly billing or $16 per month when billed annually. The comparable Industry Pro rates are $75 and $60. Team plans use per person billing and Enterprise is quoted. The hosted product encrypts data in transit and at rest, applies user access controls and supports company sign in for eligible deployments, and an independent penetration test checks its security. Organizations own their inputs and outputs, their research data never trains AI models, and they can have all of it removed at any time.

More than 1,000 GSK scientists use Undermind, and scientists at nine US Department of Energy national labs, including Brookhaven, Lawrence Berkeley, Argonne, Sandia and SLAC, use it too. Novartis, Mayo Clinic, NIH, Fraunhofer IKTS and Bain Capital are also customers, and Actipulse Neuroscience's research team uses Pro to find papers that were hard to locate. A research group could use Undermind to review a new materials question and keep a record of the methods that each relevant study used. Team members can examine specific cited passages and add documents they already have permission to use.

An assistant connected to the workspace can organize the papers and store a summary alongside the original sources. For a longer project, the group can monitor a topic and review new evidence as it arrives. The same process gives colleagues a common set of references to use when they revisit a conclusion. A research lead can inspect the source details before sharing a report or bringing the evidence into a decision. The system supports both a browser workspace and connections from other applications that a user authorizes, so a team's sources remain available for related tasks.

Vendor details

Canonical URL

https://www.undermind.ai

Category

Data analyst agent

Subcategory

Scientific literature search, analysis, and monitoring agents

Funding status

Undermind AI, Inc. is a privately held company backed by Y Combinator.

Company status

independent

Use cases & customers

Primary use cases

Run structured literature searches of many stepsRead papers and prepare evidence reports linked to their sourcesMonitor a research topic or connect a paper workspace to an existing agent

Target customers

Academic and industrial research teamsResearch operations and evidence review teamsDevelopers adding scientific research to internal assistants

Deployment options

Workspace hosted by UndermindEnterprise API and MCP integration authorized by each user

Integrations

The hosted MCP server connects research tools and user workspaces to compatible outside assistants such as Codex, Cursor and Perplexity. Enterprise customers can also integrate an Undermind research API.

In practice

A materials research group searches for a class of new findings and organizes the relevant publications to compare their methods.

An academic team assembles a research brief linked to its sources and monitors a saved topic for later papers.

An enterprise research assistant looks up approved papers from a user's Undermind workspace and saves the relevant sources and findings there.

Agentic Index coverage score

8.0 / 14 capabilities · 57%

Integrations & Tool Calling Partial

The research agent searches scholarly data, looks up papers and retrieves available open access PDFs from outside the workspace, then organizes sources and saves research artifacts in its own workspace over many steps. Its tools work on the literature and the native workspace, and writing back to outside systems is left to the connected client.

Sourceundermind.ai/mcpread 2026-11-20

Workflow Orchestration Full

A deep search can take an open objective and plan a series of queries. It examines candidate results, follows citation and author relationships, and continues while new sources prove useful, then ranks the results and ties them into a literature synthesis. Agents can also retrieve full text material and organize it inside a workspace for later questions.

Sourceundermind.ai/mcpread 2026-11-20

Knowledge Grounding & RAG Full

Workspaces contain papers, reports and source files. An authorized user can return to the same collection and search against it or retrieve full text evidence for a different question. Literature tools search available papers and retrieve citation relationships, and the product reads figures and tables within any paper the user can access.

Sourceapp.undermind.airead 2026-11-20

Human Oversight & Guardrails Partial

An outside assistant must be authorized by the person whose workspace it reaches, and that person can inspect cited evidence and review a finished research result before sharing it. The user decides which client can act and what happens to its output, with no confirmation step before each permitted edit.

Sourceundermind.ai/mcp/referenceread 2026-11-20

Security, Identity & Governance Full

Undermind encrypts data in transit and at rest and uses role based controls and multifactor authentication, with company SSO for eligible accounts, and the MCP interface authorizes against the user's own workspace. The independent assurance on offer is a penetration test, with an attestation letter customers can request under NDA, rather than a SOC 2 or ISO certification.

Sourceundermind.ai/security-one-pager.pdfread 2026-11-20

Observability & Auditability Partial

Researchers can follow a claim back to its source, read relevance explanations and review earlier results in a workspace, which gives provenance for the evidence and the final report. Workspace records show what the agent produced and the material behind it, without a trace of each research step.

Sourceundermind.airead 2026-11-20

Memory & State Persistence Partial

Agents and MCP clients work inside the user's own workspace, with earlier saved searches, reports and papers in reach across later calls and sessions, and related searches can continue from existing results. That workspace is the user's research library, which also grounds every answer, and it keeps what users and agents save there rather than a memory layer with its own scope, lifetime or purge path.

Sourceundermind.ai/mcp/referenceread 2026-11-20

Deployment & Data Residency Not documented

The web app and research API run on Undermind's own service. An organization can connect its own data, authorize outside tools with a company account and request removal of its data at any time, and Enterprise adds contract and security options. There is no region choice, private tenant or self managed agent runtime.

Sourceundermind.ai/enterpriseread 2026-11-20

Prebuilt Agents / Templates / Packs Not documented

Undermind offers one general literature research capability and a set of tools for finding, reading and organizing papers. Users bring new questions, build a collection and ask for different forms of synthesis, and the tools are operations inside the research workspace rather than distinct agent templates a customer deploys.

Sourceundermind.ai/mcpread 2026-11-20

Triggers & Channel Coverage Full

Researchers can keep monitoring a topic after an initial search and get an update when relevant publications appear, and Enterprise adds alerts on saved topics and a teammate's work. An outside client can also start an authorized search programmatically, so monitoring produces new results without a fresh prompt.

Sourceundermind.airead 2026-11-20

Model Flexibility & Routing Not documented

Plan tiers carry different resource levels for chat and deep search, and Undermind picks the models behind its own literature agent with no setting to change them. When a buyer connects an outside assistant through MCP, that client's model is the buyer's choice, but routing inside Undermind stays fixed.

Sourceundermind.ai/#pricingread 2026-11-20

APIs / SDKs / MCP Extensibility Full

Undermind's hosted MCP endpoint comes with authorization and client setup for user scoped access and a set of tools for searching, reading and organizing workspace items, and authorized tool calls can change resources in the user's workspace. Codex, Cursor and Perplexity can connect this way, and Enterprise customers also get a research API.

Sourceundermind.ai/mcp/referenceread 2026-11-20

Testing, Debugging & Optimization Partial

Researchers can inspect ranked results, compare the supporting source and judge whether a result answers their question, then refine a search or ask about a specific document. Undermind's own retrieval benchmark is a vendor run measure, and the buyer's tools center on inspecting findings, with no scoring suite to run.

Sourceundermind.airead 2026-11-20

Browser / Computer-use Partial

Undermind locates and retrieves open access papers and analyzes their contents, including figures, tables and citations, and uses files supplied to the workspace for papers outside that access. These are document fetches, with no remote desktop or graphical browser for an agent to operate.

Sourceundermind.ai/mcpread 2026-11-20

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

Industry Pro $75 per month, Academic Pro $20 per month

Subscription rates vary by segment and seat count

Free tier

What is public

Academic Pro costs $20 per month or $16 per month billed annually. Industry Pro costs $75 per month or $60 per month billed annually.

Billing mechanics

Team rates are per person. Academic Team is $18.70 a month or $15 per month billed annually. Industry Team is $75 a month or $60 per month billed annually.

Variable cost rationale

Plans are flat per person subscriptions, so cost grows with seats rather than with how many searches a researcher runs.

Additional watchouts

Free accounts have standard limits on searches and chats. Pro raises those limits and adds more capacity for full text work.

Sales call required

Mixed (some tiers require a call)

Free / trial

Free workspace with standard chat and search limits

Lowest paid plan

Industry Pro

Missing data

Enterprise pricing, API usage limits and organizational terms are set with Undermind's team.

Agentic Index verified 2026-11-20

Alternatives to Undermind

Other products that do the work Undermind does, and what each one is.

Contact us

Found a vendor we missed? Have feedback on the index? We'd love to hear from you.