Opportunities › Self-hosted AI agent workstation with control › Proposed app
BenchKeep
Run AI agents and cited document Q&A on your own servers and models
- Score
- 44
- confidence 48
- Evidence
- 6
- items on the page
- Rivals
- 2
- named in sources
- First version
- 45 d
- estimate
- Type
- Desktop app
- PWA: runs in the browser, installs on Mac and Windows
- Market since
- 2023
- Emerging market · AI estimate
ScreensConcept
Desktop app: AI agent control, document Q&A, and infrastructure management involve complex interaction and monitoring—sustained work at a workstation. Sample screens on a computer, in the proposed brand. Their numbers are sample data.
Ask
Answer checked3 sources
About this appAI proposal
BenchKeep is a self-hosted AI workstation. You run agents, ask questions across your own documents and get answers with citations, all on servers you control. You pick the model: open-source or a commercial API with your own key.
It is for developers and small teams with privacy or compliance duties, or heavy AI workloads where per-call hosted pricing hurts. Hosted assistants such as ChatGPT and NotebookLM send your data to their servers; BenchKeep keeps it where you put it.
The PWA is the control room: connect models and tools over MCP, set what each agent may touch, review its runs, and see what each model costs you. Install it on a phone or laptop and reach your own server.
- Tagline
- Your agents, your models, your servers.
- Voice
- PlainSteadyPractical
- Why this name
- A bench is where work gets done with your own tools. Keep is where you hold what is yours. Together: a workbench you keep.
In the first version
Members say which features they would use. It tells PWAS what people want; it changes no score.
Bring your own model
Connect local open-source models or commercial APIs with your own keys. Switch per agent without rewriting anything.
No member yet
Cited answers from your documents
Upload files and ask questions. Every answer links to the passages it came from, so you can check it.
No member yet
Multi-agent runs
Chain agents with roles, such as researcher and reviewer. Replay each step and see what it did.
No member yet
MCP tool connections
Add tools and data sources through MCP. Allow or deny each tool per agent.
No member yet
Brand identityAI proposal
Check the name against trademark registers before using it.
Palette
#12372FDeep Pine · primary#F6F2E9Paper · surface#1B1F1DInk · text#E0902BAmber Thread · accent#6E8F5EMoss · data
Typography
Aa
BenchKeep
ABCDEFGHIJKLMNOPQRSTUVWXYZ 0123456789
Your agents, your models, your servers.
PricingAI proposal
The material gives no competitor prices. It describes a free self-hosted tier with paid support, hosting and premium features. These amounts are starting points to test.
Self-hosted
Free
- One workspace, one user
- Bring your own models
- Cited document Q&A
- Community support
Team
$19 a month
- Per seat, up to 25 seats
- Roles and audit logs
- Multi-agent runs and skills
- Cost ledger
- Email support
Managed
$99 a month
- We host your instance
- Backups and updates
- Up to 10 seats
What it could be worthEstimate
What a software business making this much sells for today, from real deals. Pick a monthly revenue.
If it makes
Worth about
$194,000
Sale range $132,000 to $263,000
$120,000 a year in revenue
From 87 software businesses listed and sold on Empire Flippers in the last year, revenue vetted by the broker: they sell for about 19× their monthly revenue, most between 13× and 26×.
Measured on profit, the middle deal is 32× monthly net profit: about 2.7 years of profit.
An estimate, not an offer or financial advice. A real price depends on profit, growth, churn, how much of the work is the owner's, and the buyer.
How it would growAI proposal
How the product would reach new customers through its own use, and how the first ones would be found. A plan to test, not a result.
- Built-in loop
- Users invite others. Team admins invite colleagues to shared workspaces, agents and cited knowledge bases on the same server. Each invitee sees the product in daily use and can start a workspace of their own.
- Do they become customers?
- Invitees are colleagues in the same company, so they may push for a paid team plan. Test whether invited users raise the plan count, or whether one admin does all the buying.
- First customers
- Engineering leads at 10 to 50 person software or consulting firms that handle client data and want AI agents without sending files to hosted assistants.
- Where to reach them
- GitHub and open-source communities: Publish a one-command install and a guide to moving from hosted assistants. Engage with self-hosting and MCP discussions.
- Developer forums: Post worked examples of cited document Q&A on local models in self-hosting and privacy communities.
- Direct outreach: Message engineering leads at small firms with client confidentiality duties and offer a setup call.
- First milestone
- Five teams pay for the Team plan after a 30-day trial on their own servers.
Why people already pay for this
- VerifiedGitHub repository gained 306 stars in 167 days
- Public claimSourceWeft is an open-source, self-hostable alternative designed to compete with hosted AI services like ChatGPT and NotebookLM.
- Public claimThe platform enables users to maintain ownership and control of their infrastructure and models rather than relying on vendor services.
- AI inferenceThe project has significant adoption and interest, with 306 GitHub stars and 25 forks, indicating market validation for the self-hosted AI agent category.
Tech stack
Every proposal builds on the same PWA base, installable from the browser on any phone or computer. Dashed boxes are what this app adds.
- AppNext.js and TypeScript
- PWAManifest and service worker
- UITailwind CSS
- Data and authSupabase Postgres
- PaymentsStripe
- EmailResend
- HostingVercel
- This app · Self-host packagingContainer image with compose fileCustomers run the server on their own machines, so install must be one command.
- This app · Vector searchpgvector in PostgresCited retrieval without adding a separate database to host.
RisksAI inference
- CommercialThe evidence shows interest in a similar open-source project (306 stars in 167 days). Stars may not become paying teams, and a free self-hosted tier may satisfy most users.
- TechnicalSelf-hosting varies widely: GPUs, local model runtimes, and upgrades. Support cost per customer could be high for a small team.
- AILocal open-source models may give weaker cited answers than hosted ones, which would hurt comparisons with ChatGPT and NotebookLM.
- PlatformMCP and agent standards are still changing. Tool connections may need frequent updates.
- Intellectual propertyThe signal names an existing open-source project in this space. Licences of any reused code and the name need checking.
Everything marked AI proposal or AI inference was written by a model from the evidence on the opportunity page. The proposed name must be checked against trademark registers before it is used.