Turn 40 messy documents into a decision you can defend.
Local-first desktop app · macOS at launch · your keys, your machine
Turn 40 messy documents into a decision you can defend.
Korvo is a private research workspace that runs on your machine. Drop in decks, PDFs, notes, and transcripts - it organizes the context, plans the analysis, cites material claims, and keeps a record of how you decided.
Trace every decisionTwo models, then a critique passWorkspace data stays local
Download free for macOSSee a project end to end →
Explorer is free · Other platform availability is listed on the download page
One project, end to end
From 12 scattered files to an investment decision.
Capture → Plan → Verify → Challenge → Decide Scroll to follow the project →
Capture01 / 05
Bring the whole deal into one project
12 files: pitch deck, financial model, founder notes, market research, and two call transcripts.
korvo · Series B Diligence · Context
Files
pitch-deck.pdf
financials.xlsx
founder-notes.md
call-01.txt
Project library
12 sources · indexed
pitch-deck.pdf2.4 MB
financials.xlsx880 KB
founder-notes.md41 KB
call-01.txt128 KB
call-02.txt96 KB
\+ 7 morelocal
Plan02 / 05
Review the analysis before it runs
Korvo detects the goal, makes assumptions visible, flags missing context, and proposes the work.
korvo · Series B Diligence · Plan
Detected goal
Evaluate Series B opportunity — unit economics, retention, comps
01Analyze unit economicspending
02Test retention assumptionspending
03Compare portfolio benchmarkspending
Missing: cohort churn · cap table
Approve plan
Verify03 / 05
Trace an answer to the source line
Every material claim links back to the evidence selected from the project.
korvo · Citation inspector
Burn multiple is 2.4× on trailing twelve months.
grounded
“Net burn: $1.2m · Net new ARR: $500k · TTM”
financials.xlsx·FY26 model · Sheet “Unit econ” · R14:C6
Open source in workspace → jump to selected span
Challenge04 / 05
See where two models diverge
One model builds the case. A second critiques it. The revision and unresolved questions stay visible.
korvo · Two-model critique · beta
Thesis
Model A
Efficient growth and 2.4× burn support the round at the ask.
Critique
Model B
Retention evidence is cohort-incomplete; concentration risk underweighted.
Revision
Merged
Conditional proceed — pending churn diligence and top-10 customer check.
Decide05 / 05
Preserve what you decided and why
The conclusion, evidence, caveats, model details, and timestamp remain together for later review.
korvo · Decision history · audit trail
Proceed to partner review
30 Jul 2026 · 14:22
Reason: strong unit economics, subject to validating gross retention and customer concentration.
Models · A + B critique
Sources · 6 cited
Open Qs · 2
Export · PDF / MD
What Korvo is
Not a chatbot. Not just a workspace.
A decision engine.Chat gives you an answer. Korvo helps you define the job before the answer is written.
A workspace stores your material. Korvo connects each conclusion back to the material that supports it.
A decision engine keeps the plan, challenge, evidence, and final call together so another person can inspect them later.
The workflow
Capture → Structure → Plan → Execute → Preserve
1
Capture
Drop files, notes, and pasted text into one project. Keep the messy source material together.
2
Structure
Korvo indexes the project locally so you can search and retrieve the right context.
3
Plan
It identifies the goal, assumptions, missing context, and analysis steps. You approve the plan before execution.
4
Execute
Produce briefs and syntheses grounded in evidence selected from your workspace, with citations back to the source.
5
Preserve
Keep the conclusion, source links, model details, revisions, and unresolved questions together.
Built for
People who have to show their work.
Investors
Run diligence across decks, financials, founder notes, calls, and market research. Pressure-test the case before it reaches partner review.
Founders
Turn board material, customer evidence, and scattered strategy notes into a plan with explicit assumptions.
Operators
Synthesize reports and project context into an actionable brief without losing the evidence behind it.
Researchers
Organize a literature set, compare claims, and produce cited synthesis from the source material.
Privacy and models
Local-first workspace. Transparent model traffic.
Your workspace, files, local index, and decision history stay on your machine. When using a cloud model, selected context goes directly to your provider under your own key and agreement. Or connect Ollama and nothing leaves your machine at all.
Local-first architecture
Files, the local SQLite index, notes, plans, and decision history reside strictly on your device. Korvo never holds your workspace data on its servers.
BYOK cloud traffic transparency
Connect OpenAI, Anthropic, or Google directly with your own key. Prompts and selected context go to that provider under their terms. Korvo never proxies, inspects, or logs them.
100% on-device option
Connect Ollama for full offline operation on local models. No prompt, project context, or document data leaves your machine.
Data rights, zero lock-in
Export outputs to PDF and Markdown. Delete local workspace data anytime. Your files, notes, and generated outputs remain yours.
Medha
Models generate. Medha measures. You decide.
Medha is Korvo’s on-device quality-measurement layer. It tests retrieval with deterministic metrics such as recall, MRR, and nDCG without model calls or private corpus content leaving the device. It does not generate answers or claim that an answer is true.
<10ms per operation$0 per evaluationZero network callsHuman-labelled release evaluation in progress
Pricing
Prove it on one real project for free.
Korvo does not resell model tokens. Cloud-provider API usage is billed separately by the provider you choose.
Explorer
$0forever
Enough room to test the full workflow on a meaningful project.
- 1 active project
- 15 files or research papers
- 20 notes · 1 table
- Indexing and grounded chat
- Plan generation
- 1 saved output
- All AI providers (BYOK)
- Unlimited projects, files, notes, and tables
- Two-model critique beta
- Unlimited saved outputs
- Export to PDF and Markdown
- Decision history and source links
- Priority support
Early access
Pro
$149/year
Renews at $149/year while the subscription stays active · planned public price $299/year
Unlimited projects and the complete decision workflow, including the experimental two-model critique.
Download free - sign in with GoogleOr upgrade to Pro - $149/year →
One honest cost comparison
Korvo can consolidate project notes, local document retrieval, citations, and decision records. It does not eliminate model cost: cloud AI usage is charged separately to your API account and varies by provider and usage.
Questions before you download
The practical details.
Do I need my own API keys?+
Yes for cloud models. Add a key from OpenAI, Anthropic, or Google and requests go directly from your device to that provider. API usage is separate from Korvo Pro; a practical individual-use budget is often about $5–$30 per month, but the actual amount depends on the model, document volume, and how often you run multi-model critique.
What actually leaves my machine?+
Your workspace files, notes, local SQLite index, plans, and decision history reside strictly on your device and are never uploaded to Korvo servers. Cloud-model prompts and selected context go directly to your chosen provider under your own key and agreement. When you explicitly run an MCP tool, its arguments and required authentication go directly to the server you connected. Those requests are governed by the provider or server’s terms; Korvo does not proxy, log, or inspect them.
What if I do not want to use a cloud model?+
Connect a local model through Ollama. Model requests then stay on your machine, so no prompt or project context is sent to a cloud-model provider. Model quality and hardware requirements depend on the model you choose.
Is my work locked in?+
Your workspace is stored on your device and remains yours. Pro can export outputs to PDF and Markdown, and you can delete local workspace data at any time. Korvo does not claim rights over your files, notes, or generated outputs.
Can Korvo connect to MCP tools?+
Yes. Connect the built-in Superhuman Mail preset or add a custom MCP server under Settings → Integrations → AI Tools. You can inspect discovered tools and run them manually with JSON arguments. Chat models cannot autonomously execute MCP tools; mutating or unannotated tools require an Allow once confirmation.
Current status
Early access
Useful now. Still being evaluated honestly.
The two-model critique remains beta while real-project dogfood and blinded evaluation continue. Medha’s synthetic baselines are available; human-labelled release evaluation is still in progress. Access is rolling out in small batches, with early-access delivery targeted within seven days of purchase.
Bring the messy project.
Leave with a decision you can explain.Start free with one project and up to 15 files.
Download free - sign in with GoogleOr upgrade to Pro - $149/year →