Hacker News: anuptalwalkar

New comment by anuptalwalkar in "Tesla Roadster"

anuptalwalkar — Fri, 16 Oct 2026 06:17:51 +0000

I have no words. Its like Elon asked Grok where can I take my business, and it suggested combine rockets and cars.

New comment by anuptalwalkar in "Show HN: AgentDrive – persistent, versioned file storage for AI agents"

anuptalwalkar — Tue, 15 Sep 2026 01:05:10 +0000

Thanks for the reply. would that be against some GDPR policy, not that should be a concern in the beginning, but something to think about 'wipeouts'.

I struggled with deletion and clean ups a lot and found that is the hardest part to solve of it all.

New comment by anuptalwalkar in "Show HN: AgentDrive – persistent, versioned file storage for AI agents"

anuptalwalkar — Mon, 14 Sep 2026 22:03:43 +0000

A few questions - Are these read only? how do you relay ownership? If a rogue agent goes and deletes files for example, how would the system treats that error?

I recently open sourced Recall https://github.com/Polign/recall, which is sharing the memory and typed correctness across agents.

New comment by anuptalwalkar in "Show HN: Typed agent memory with corrections and history"

anuptalwalkar — Mon, 14 Sep 2026 18:49:32 +0000

I wrote this with a primary goal in mind to share facts and context across Agents.

It is open source and the way this works is by defining the typed memory around default predicates ex -https://github.com/Polign/recall/blob/main/internal/engine/d...

The predicates can be customized for individual use cases, and typed memory will extrapolate from the conversations you have with it. For example, in some use cases you should be able to deterministically separate actions from decisions. The agent, instead of guessing or hallucinating, will be able to successfully recall the precise decisions, and whether the actions are permitted or not.

Show HN: Typed agent memory with corrections and history

anuptalwalkar — Mon, 14 Sep 2026 18:38:31 +0000

Article URL: https://github.com/Polign/recall

Comments URL: https://news.ycombinator.com/item?id=49701788

Points: 1

# Comments: 1

New comment by anuptalwalkar in "Show HN: Vector transfer project, replicate vectors/metadata across databases"

anuptalwalkar — Thu, 10 Sep 2026 22:34:42 +0000

I got tired of setting up various vectorDBs for performance and benchmarking, so I wrote this universal transfer setup to move and replicate vectors and metadata across databases.

It runs locally. Through UI, there is also a way to connect your host to a control plane if you want to visualize and monitor the transfer, but that part is optional. I found it useful to track and monitor all my previous transfers.

The tool currently supports a variety of connectors. If something is missing, connectors are easy to add here: https://github.com/Polign/vector-transfer/blob/main/CONTRIBU...

I would love to have some feedback, or any features/performance stuff worth updating.

Show HN: Vector transfer project, replicate vectors/metadata across databases

anuptalwalkar — Thu, 10 Sep 2026 22:31:44 +0000

Article URL: https://github.com/Polign/vector-transfer

Comments URL: https://news.ycombinator.com/item?id=49651027

Points: 3

# Comments: 1

Cost of running agent memory/vector search

anuptalwalkar — Mon, 07 Sep 2026 02:04:29 +0000

Article URL: https://polign.com/agent-memory-at-scale

Comments URL: https://news.ycombinator.com/item?id=49593006

Points: 2

# Comments: 0

New comment by anuptalwalkar in "Show HN: Personal Context MCP"

anuptalwalkar — Sun, 06 Sep 2026 03:22:30 +0000

Interesting concept. Do you have some demos of any use cases?

Curious why you chose github for memory.

New comment by anuptalwalkar in "How is AWS console so broken?"

anuptalwalkar — Wed, 02 Sep 2026 05:02:13 +0000

I hear you. It at least works in chrome, being Google after all.

I would imagine someone would let gemini or whatever aws is using to run wild and say 'fix everything in the UI'. It wouldn't be the worst idea of the year.

How is AWS console so broken?

anuptalwalkar — Wed, 02 Sep 2026 03:45:54 +0000

A small rant, after painful user experience. I admit I visited it after a while spending on GCP space, but the console looks like it is either not maintained at all, or individual teams just throwing stuff at it and causing a mess.

My simple workflow - I want to copy millions of objects from one bucket in us-east-1 to us-west-1

1. I can kick off the replication rule. but oh wait, there is no monitoring, unless you select it in edit rules. Even then, the experience is broken and I have now no idea if anything is happening in the background.

2. There is batch operation. It doesn't work. If I select source region for aws console, I can't see destination region bucket to execute copy batch. Vice-a-versa, if I select destination region, I can't select any buckets from the source region.

I can perhaps use a CLI, which is a long list of instructions to follow and execute without pulling my hair out.

I asked claude code to just figure out out to create a job and execute it and tell me the state. It didn't even hesitate.

Comments URL: https://news.ycombinator.com/item?id=49531472

Points: 1

# Comments: 3

New comment by anuptalwalkar in "Show HN: A lightweight, stateless database for agent memory"

anuptalwalkar — Mon, 31 Aug 2026 18:36:03 +0000

Yes, its likely the keyword search problem. I had decent experience (including benchmark and recall testing) to use the hybrid search. You can check my demo - https://demo.polign.com which has about 12.5 million wikipedia passages working with hybrid search.

New comment by anuptalwalkar in "Show HN: A lightweight, stateless database for agent memory"

anuptalwalkar — Thu, 27 Aug 2026 23:59:23 +0000

The policy layer you’re describing lives in the harness, and yes, it changes over time. What I wanted to showcase is polign_db underneath with typed records, hybrid index, attached to your bucket. You can swap your memory design without migrating the storage or spawn nodes to access it quickly.

I see your point on observability. I haven’t baked it into polign yet, but I have been thinking about ‘how’ I would do it a very user friendly way. Glad to have a conversation about it.

New comment by anuptalwalkar in "Small Models Have Arrived"

anuptalwalkar — Thu, 27 Aug 2026 21:55:30 +0000

I kind of agree with your assessment. Running models not just on local, but cheap and lightweight frameworks will drive the next phase.

Not trying to plug, but I do't know any other way. I wrote a piece couple of days ago on small models and memory usage on the edge devices- https://polign.com/blog-edge-agent-memory and https://news.ycombinator.com/item?id=49450816 closing on the same problem.

New comment by anuptalwalkar in "Show HN: A lightweight, stateless database for agent memory"

anuptalwalkar — Thu, 27 Aug 2026 19:26:54 +0000

Thanks, is the concern being non-starter on the trust side? the binary makes zero outbound calls. It talks to your bucket and nothing else, and that's verifiable with egress monitoring. I'm putting up a security page that spells it out.

Open-sourcing parts of it isn't off the table. I kept the core closed for now to understand the interest in it.

New comment by anuptalwalkar in "Show HN: A lightweight, stateless database for agent memory"

anuptalwalkar — Thu, 27 Aug 2026 19:14:13 +0000

When I started, I was using what was available out in the world. The vector dbs I used and even the first version I built weren't really optimized to use RAM and resources conservatively. One of the aspects of Polign is it's optimized towards cold queries and only warms the cache when needed, all while keeping the recall accuracy in acceptable range. I accepted slower first-query latency on cold data as the tradeoff. For example, the demo serves 12.5 million passages from S3 with the server idling around 37 MiB RSS (I realized there is a 20 MB gcp sdk in there).

Hosted db have query costs that grow with the corpus, and self hosted have operational overheads. They do add up over time and I wanted cost flat, which took a while to get it right.

If a vibe-coded version covers your use case, you don't need this. I have added my own experience working and optimizing the storage into this, and Im certain it holds up well for a growing corpus and with simplicity to use it across multiple readers.

New comment by anuptalwalkar in "Show HN: A lightweight, stateless database for agent memory"

anuptalwalkar — Thu, 27 Aug 2026 18:02:01 +0000

It's wordplay. I wanted it to be lightweight like pollen, but carrying all the information you need to build on top of it. The 'align' part nods to the core vector search aligning with your queries.

New comment by anuptalwalkar in "Show HN: A lightweight, stateless database for agent memory"

anuptalwalkar — Thu, 27 Aug 2026 05:00:16 +0000

Thanks for taking a look. Yes, I kept it closed source for now. I'll mention that on the page.

Litestream with SQLite would solve some of these use cases, and backing up to S3 or GCS works well.

Here I was focused on more dynamic workloads. For example, larger indexes and corpus that can be accessed by multiple readers. I didn't want to invest in local storage, so the data is made to stay cold until it's actually accessed into warm cache.

Also, one node can serve all of that in one collection while serving typed agent memory in another.

Perhaps I missed something, but storage, memory, and operational overhead is something I wanted to worry less about.

Show HN: A lightweight, stateless database for agent memory

anuptalwalkar — Wed, 26 Aug 2026 15:24:51 +0000

I've been working on Polign and built a small prototype around something I've been thinking about with agent memory. I have built a lightweight/stateless vector db + BM25 search which works really well with typed facts and structured queries.

It uses your own S3, or GCS bucket as primary storage, and restarting a node is fairly quick.

Demo + writeup: https://polign.com/blog-edge-agent-memory

Live search demo: https://demo.polign.com

Docs: https://polign.com

Comments URL: https://news.ycombinator.com/item?id=49450816

Points: 36

# Comments: 14