Same model, a different proposal.Professional-grade AI for researchers.
Upload your notes and prior papers and get an R&D proposal in the call's own form, every reference checked against the original.
Search chats and places
Redrob Cowork
RedrobConnection secureWhat should we build today?
Redrob AutoRedrob
Claude Opus 5.5Anthropic
Gemini 4 ArgonGoogle
Muse Spark 1.2MetaWork that takes expertise is what Redrob does best
Redrob is AI and can make mistakes. Check its work.
Why one AI model can write proposals of very different quality
Finding and reading the papers, opening the call's form and checking citations happen outside the model, in its setup.With that setup in place, the same model writes a proposal fit to submit.
- Web chatBenchmark
- Desktop appAbout +8%
- Coding agent appAbout +19%
- + Skills built for the taskAbout +27%
- + Two or three of your firm's own templatesAbout +48%
The difference in the work comes from the setup the AI model runs in.
Read the manualFor researchers who need professional-grade AI
From a company lab's project proposal to a university paper: built for writing where one source can decide the review.
- Small business researchers
- Large company researchers
- Venture company researchers
- Full-time university faculty
- Mid-size company researchers
- University lecturers
- Research and contract professors
- Public institute researchers
- Adjunct professors
- Visiting professors
- Professors and researchers
- Management consultants
- VC and PE investors
- Doctors
- Lawyers
- Patent attorneys
- Labor attorneys
- Accountants
- Tax accountants
- Appraisers
Just ask as you normally would.A developer setup finds the prior work.
No coding needed. A developer setup searches KCI, RISS and NTIS for prior studies and similar projects, reads them and checks each citation against the original. Each research skill was built over 100+ hours from hundreds of reference documents.
- Research paper draftingDrafts a paper section by section from your own sources
- Research source finderFinds and checks real studies, so no citation is made up
- Natural academic writingTakes the AI-sounding phrasing out of Korean papers
Your participants' raw data is not used to train AI
Contact details and ID numbers in survey answers and interview notes are masked on your computer before anything is sent. Raw data files stay locked in Vault.
The AI Firewall now runs on your device
Before your question even reaches an AI, personal details are de-identified right on your computer. The AI never sees them, and the details are filled back in when the answer comes back.
Your documents, safe in Vault
Templates and documents live in Vault, your cloud storage, and only come out for the work that needs them.
- End-to-end encryptedDocuments are locked on your device before they are stored. Only you and the people you allow can open them.
- Up to 100 GB each, about 1 million pagesOn Pro and Team that's about 1 million pages of contracts, ledgers and scans, roughly 2,000 reams of paper. Free comes with 1 GB, Plus with 20 GB.
- Never used for trainingWhat you type and upload is never used to train AI models.
The desktop app adds another layer of protection
You can start on the web. For confidential documents, use the desktop app.
- Hidden on your computer firstAI Firewall runs in the desktop app, so personal details are hidden before a question leaves your computer.
- Admins lock the levelA team admin can lock one AI Firewall level for everyone, so every member is covered the same way.
When a stronger model for reading papers comes out,Redrob Auto keeps up.
Spend the time you'd spend comparing new models on research. It picks the right model for every request.
- 1Claude Opus 5.5AnthropicDeep reasoning, careful writing. Builds a solid case for the proposal.9.6Job benchmark
- 2Gemini 4 ArgonGoogleReads long papers in one pass. Groups dozens of prior studies by question.9.5Job benchmark
- 3GPT-6 AstraOpenAIStrong on multi-step work. From literature search to write-up in one go.9.4Job benchmark
- 4Muse Spark 1.2MetaQuick and careful. Gets every reference entry and its style right.9.2Job benchmark
- 5Claude Opus 4.8AnthropicProven and steady. Trusted with yearly reports and replies to reviewers.9.0Job benchmark
Learn more about Redrob
Answers to the five questions we hear most.
What is Redrob?
An AI agent for people whose work ends in a document, like lawyers, accountants, investors and doctors. It wraps the same AI model in skills for your field, your firm's templates and a workspace built for the job, so you get work a senior can review as is. Start on the web, then finish in your firm's format in the desktop app.
How does AI Firewall work?
Before your question reaches an AI, the desktop app swaps personal details for stand-ins like "Email 1" on your computer, then fills them back in when the answer arrives. Standard hides emails, phone numbers, resident registration numbers and card numbers. High also finds names of people, companies and places (it can miss some). Strict also turns voice off.
If the model is the same, why is the work different?
Because results depend a lot on how the model is run, not just on the model. Public benchmarks show the same model scoring higher in a tooled setup like a desktop app or a coding agent app. In Redrob's internal evaluation, compared with web chat, task-built skills made work about 27% better, and adding two or three of your firm's own templates made it about 48% better.
Do I need to know how to code?
Not at all. Type what you need in plain words, and Redrob Code does the developer-style work behind the desktop app: reading files, building the document and checking the result. No commands, no setup.
How much does it cost?
Free is free, Plus is $10 a month, Pro is $25 a month and Team is $30 per seat a month. On any plan you can top up credit from $5. See the pricing page for the details.
The proposal you put before reviewers,written on Redrob.
Start from an R&D proposal in the call's own form, with every reference checked.