@ai.honeycove: Tencent DB Agent Memory just open-sourced a local AI memory system that mimics human long-term recall. It boosts AI agent efficiency by cutting token costs 61% and improving task success over 50% with layered, evolving memory that remembers your preferences and habits. #tencentdbagent #aimemory #openaiinnovation #agenticai #localai
Claude already have long term memory, just need to turn it on
2026-08-13 04:06:32
3
blackyskeef :
ohhh it's tencent😭😭😭😭😭😭😭😭 we are fried
2026-08-15 12:32:54
0
Bill :
I may have some knowledge twisted, but how does it access the memory without pulling into context? I thought that was one of the challenges because it would need to train on the ‘remembered’ data to actually be a part of the model. Otherwise this is another way of strategically accessing what gets pulled into context. Still helpful, just trying to understand.
2026-08-12 14:54:40
3
jules :
cool list. workbeaver has been interesting to try, especially when i have repetitive tasks that are easier to explain than manually set up.
2026-08-14 18:10:06
60
Qadenza :
Lots of similar stuff like this exists but if it's easier to set up properly it could be popular, especially if it were easy to use it with multiple providers or qadenza.ai (shameless plug)
2026-10-02 03:50:53
0
Jeb Jarrell :
stick with ob1, tencent is too new and burdensome
2026-08-12 12:30:46
3
Claw Nasty :
Isn’t obsidian better because you can share it between multiple systems?
2026-09-27 17:12:46
0
ebokianknight :
This is also just unverified hype. Models need to spend context learning to use a tool they aren’t trained on. And the context still needs to be added to affect the models document. Is also has no conflict resolution or way to handle stale data. Rushing for hype over substance
2026-08-13 10:57:21
2
Arsenicx2 :
Been doing this in Open Claw for months now 😂
2026-08-13 10:22:43
2
teknowledgy :
There's already multiple open source versions of this, like Hermes. Has been for awhile.
2026-08-13 06:03:17
2
Jives :
is it just running RAG?
2026-08-12 23:01:06
2
science and technology (Elie) :
i linked my Ai to Google drive and asked him to back up any useful information to it so any new ai can pick up from there ,i know its less efficient than a Db but itvwas simple and implemented it longbtime ago 😳😳
2026-09-08 11:08:36
1
Ayase :
obsidian?
2026-08-12 15:43:48
5
capet :
Claude Recall HQ was doing this long before Anthropic was doing it
2026-08-13 04:52:37
1
llllllllllll :
I don't want to use my phone number
2026-09-10 19:34:24
0
Reggie Raccoon :
I've been doing this since gpt 4.o by exporting sessions as JSON
2026-08-13 14:04:01
0
Quantfit :
But Hermes does this and obsidian second brain so is this suppose to be new
2026-08-16 13:12:22
0
Garnie Bolling :
great stuff
also you can find Obsidian + Claude
that works with memory by leveraging indexes for each of your assets within Obsidian, so not just claud memories, to make Claude more efficient
2026-08-13 15:12:11
0
Z :
yeah this is just rag lol. been around for years at this point
2026-08-13 17:25:55
0
Luza Maharjan :
tencent stole my idea bro, not fair
2026-10-22 09:30:01
0
Victor :
thats just hermes doing it natively
2026-08-15 10:51:45
0
gringo351 :
use GetShit Done
2026-08-14 14:43:02
0
chunky fresh :
so is the session memory lost in time
2026-08-13 19:24:22
0
Apdiqafaar 🏝️ :
but model is still 1m context
2026-08-13 14:42:44
0
To see more videos from user @ai.honeycove, please go to the Tikwm
homepage.