Tilores entity resolution
One trusted view of every customer, company, transaction
Start new thread

Tilores Studio desktop entity resolution - Entity resolution on your machine. No cloud, no signup.

by•
Entity resolution has always been enterprise software: book a demo, sign a big contract before you see it run on your data. Tilores Studio changes that. The same real-time matching engine behind our cloud product, running entirely on your machine. Load your CSV, resolve duplicates live, nothing leaves your laptop. Now a local MCP server lets Claude Code, Codex and other AI assistants search, import and steer Studio against your data. Free up to 100k records. macOS, Windows, Linux.

Add a comment

Replies

Best
Maker
📌
I'm Steven, one of the co-founders of Tilores. Entity resolution (working out that "J. Smith", "John Smith" and "Jon Smyth" are the same person across messy datasets) has always been enterprise software. Book a demo, talk to sales, contract, months of procurement before you ever see it run on your own data. Which is backwards, because the teams who need it most (banks, compliance, fraud, healthcare) are exactly the ones who can't hand their data to a stranger's cloud just to evaluate a tool. So we built Tilores Studio. It's the same real-time matching engine that runs our cloud product, packaged to run entirely on your own machine. Download it, drop in your own CSV, and watch it resolve duplicates in real time. Nothing leaves your laptop. It's free up to 100,000 records, enough to test it properly on real data rather than a toy sample. It ships with two pre-configured use cases (people and companies), sample datasets, a golden-record view, an entity graph, and the same GraphQL API as our cloud product, so if you outgrow it there's no migration. New in this build: Studio runs a local MCP server. So AI assistants like Claude Code and Codex can now drive it directly, searching your entities, exploring matches, importing and exporting data, and steering the app, all against your local data with nothing leaving your machine. If you're building agentic workflows over messy data, this gives your assistant a resolved entity layer it can actually query. We built this because we were tired of telling curious engineers "book a demo" when they just wanted to try the thing. Now you can. Happy to get into anything: the matching rules, the MCP integration, where it breaks, what's on the roadmap. I'll be here all day.

 congrats Steven and team!

Congrats team for shipping🙌 leveraging a local mcp setup to explore and sort matches without writing massive python scripts is pure leverage. qq does the local server support simultaneous concurrent client connections if we have both cursor and claude code hitting the database at the same time?

Maker

 yes, concurrent data access is not an issue for most of the tools the server provides. There are a few tools though that wouldn't make sense to use concurrently: namely everthing that drives the UI state (switch pages, open diagram, etc.). But everything data related is safe.

Really cool - and I an definitely see this being handy as you deal with entity resolution. A problem I've run into many times in the past - "how many different variants of JP Morgan exist" - many more than you think!

Great to see!

Maker

 Glad to hear that you like it. As a matter of fact, we have quite a sophisticated and well working approach for company name matching (see Exiger case study on our website). Let us know if you're struggling next time, we might be able to help.

Congratulations on the launch. Looks amazing!

Maker

 Thanks Jan! Appreciate the feedback.

honestly the speed is what got me, like linking messy records across a few csvs in basically seconds. pretty handy if your stack is full of half-synced data.

Maker

 If you like the speed of the Studio, you would love the speed of the SaaS solution. We can easily scale into tens of thousands of records concurrently.

Entity resolution usually loses buyers at the data-sharing step, so running it fully local removes the biggest procurement blocker for regulated teams. The open question is match accuracy without cloud-scale reference data. How are you handling fuzzy matches and dedup thresholds on a single machine? That determines whether this replaces a pipeline or just supplements one.

Maker

 Currently the Tilores Studio uses optimized fixed rule sets for the two built-in use cases (person and company matching). While they are a good start to test the general process on customer side with the least efford, we typically fine tune the matching process together with the customers needs during onboarding or evaluation periods to get the best results. The result of that is a solution that can be either hosted by the customer themself or by us and can easily be implemented in existing pipeline. I'd recommend to look at a few of the use case studies on our website that provide further details.
We're planning to also offer customization options in the Tilores Studio directly. Expect further improvements comming soon. :)

Congrats on the launch. This is so needed, I work in an environment where we cannot upload our data to any vendor.

What are my options beyond 100k records? Is there an on-prem or bring-your-own-cloud version of Tilores as well?

Maker

 primarily our production deployments we run on AWS, but we can also do on-prem (which would mean we can deploy on any cloud) via a container. Just need to be able to handle the Devops side of things. Let's have a chat after the launch to discuss your use case.

I like that you're exposing the entity resolution layer through MCP. Agents are only as reliable as the data they can access, and resolving duplicate records before they reason over them seems like a solid architectural approach.

Maker

 True. We see a huge benefit for LLMs to not judge on the data themself, but to first unify it and then make it accessible. This provides accurate context for LLMs in various use cases.

Super useful this. I work with deploying ai agents across sales and marketing teams. Making them effective often means harmonising salesforce data on the sales side and hubspot and other systems that marketing uses. A lot of problems I didn’t think were problems and it got me into the entity resolution rabbit hole.

Maker

 Rabbit hole fits very well for this kind of challenge. It's always interesting to see how people completely misjudge how complex it can get. Well, I guess there is a reason that despite being an 80 year old problem, there is no perfect solution yet for ER. Or should I say "was". ;)

Huge congrats on launching Tilores Studio! Bringing enterprise-grade entity resolution completely local via a desktop app is a massive win for privacy-conscious teams who can't ship data to third-party clouds.

Since entity resolution can be quite heavy on system resources, what are the recommended local hardware specs when processing close to the 100k record limit? Also, how does the local engine handle memory allocation during massive deduplication tasks?

Maker

 pretty much any potato should be able to handle 100k records - challenge usually starts at 1M or more records for badly optimized algorithms. Our SaaS/on-prem versions can easily handle hundreds of million records - by default they are built on a serverless stack and as such users wouldn't have to worry about machine requirements.

12
Next