pranabsarkar

6 ms·

1.
by pranabsarkar 1mo ago
For memories I use YantrikDB MCP server which connects all my workspaces among distributes systems. To collaborate runtime with workspaces I use https://github.com/spranab/swarmcode
2.
by pranabsarkar 1mo ago
I am working on AI memory. Have been working on YantrikDB since January this year. Started as a pet project solving my own problem but now its growing. It's fun and satisfying at the same time
3.
by pranabsarkar 4mo ago
Thank you, Sir!
4.
by pranabsarkar 4mo ago
OP here. I am terrible at marketing so I started to build a thing for people like me: I Can't Market. The reason is after getting Claude Code/Copilot subscriptions I am working on many projects in my spare time. I am using many of
5.
Show HN: icantmarket – A verified help exchange for devs who can't market(icantmarket.com)
2 points by pranabsarkar 4mo ago|4 comments
6.
by pranabsarkar 4mo ago
I have been using personally for last 4 months in different modes: standalone, mcp, clustered. The way I communicate with AI agents is now fundamentally different. Since the agent has full memory the conversation is more coherent and I don&
7.
Show HN: YantrikDB – persistent memory for AI agents(github.com)
1 points by pranabsarkar 4mo ago|1 comments
8.
by pranabsarkar 5mo ago
Let me know your findings.
9.
by pranabsarkar 5mo ago
Fixed eval — 80 tools, 200 queries, 4 model sizes. +10pp came from "all tools" vs "tiered" on 1.5B. You're right about stability. Haven't run rotated/rephrased evals yet. The 89% baseline (when models knew
10.
Show HN: Tier – Adaptive tool routing that makes small LLMs 10pt more accurate(github.com)
4 points by pranabsarkar 5mo ago|3 comments
11.
by pranabsarkar 5mo ago
Supplements, definitely — for a specific workload. General document retrieval at scale (millions of chunks, read-heavy, doc-search patterns) is well-served by existing vector stores; YantrikDB doesn't compete on throughput. Where it&#x
12.
by pranabsarkar 5mo ago
I have been using the memory while building it. I have a central server and all my workspaces are connected to it via the MCP server. This changed everything for me. But that's me. Now I don't have to repeat things, the agent know
13.
by pranabsarkar 5mo ago
I am using this while developing and found it very useful to me. since all of my workspaces are connected it has knows all about myself and my infra. Also now we have a bonding and I can do great conversations. So decided to convert the sta
14.
by pranabsarkar 5mo ago
You're right. Pruning to isolated facts loses the structure that disambiguates them. Three partial mechanisms the system has, none of which fully solve your point: Graph edges carry scope. Alice ceo_of Acme and Andy ceo_of Amazon are t
15.
by pranabsarkar 5mo ago
Fair criticism — and the failure modes you describe aren't mem0-specific, they hit any system that extracts atomic facts from conversation. I hit a couple of them today while benchmarking YantrikDB's own consolidation (see my repl
16.
by pranabsarkar 5mo ago
Update — ran a real bench on the live cluster (59 memories: 8 canonical facts × 3-4 paraphrases + 6 seeded contradictions + 20 distractors). Numbers: duplicates per query (top-10): 0.9 → 0.0 top-result correct: 75% → 87.5% 11 consolidations
17.
by pranabsarkar 5mo ago
Thanks, really appreciate it. I am using the server as MCP server and connected all my workspaces. It has definitely changed my experience.
18.
by pranabsarkar 5mo ago
Thanks for the careful read — the "schema is ballooning" observation is real and I've felt it building this. You're pointing at a genuine design tension. My counter, qualified: deterministic consolidation is cheap and re
19.
by pranabsarkar 5mo ago
Fair. "Does consolidation actually improve recall quality on a running system?" is exactly the benchmark I haven't published, and it's the one that would settle the question. What I do have right now: 1178 core unit test
20.
by pranabsarkar 5mo ago
No LLM in the loop. The consolidation pass is deterministic: Pull the N most recent active memories (default 30) with embeddings Pairwise cosine similarity, threshold 0.85 For each similar pair, check if they share extracted entities Shared
21.
by pranabsarkar 5mo ago
Fair question. Supermemory is a hosted SaaS built around embedding + ranking. YantrikDB is self-hosted and adds three things Supermemory doesn't do as first-class operations: think() — consolidates similar memories into canonical ones
22.
by pranabsarkar 5mo ago
Author here. I built this because I was using ChromaDB for an AI agent's memory and recall quality went to garbage at ~5k memories. The agent kept recalling outdated facts, contradicting itself across sessions, and the context window w
23.
Show HN: A memory database that forgets, consolidates, and detects contradiction(github.com)
48 points by pranabsarkar 5mo ago|33 comments
24.
by pranabsarkar 5mo ago
That is definitely true, no denying in that. To be honest I was experimenting something for fun.
25.
A 24/7 live AI-generated sitcom where agents write their own episodes(tv.bothn.com)
2 points by pranabsarkar 5mo ago|2 comments