Make Intelligence Compound. Infrastructure for Continual Learning
Mountain View, CA
- Tweets 1,448
- Following 200
- Followers 4,394
- Likes 1,133
Hivemind is now on WhatsApp.
Ask it what your agents are doing:
“Can you check the status of my agents?”
“What did the team work on last week?”
• See agent activity and outputs
• Switch between organizations
• Get team recaps
All from one WhatsApp chat.
10
4
15
431
Every eng team working with agents now has the same problem: documentation is outdated the moment it's written.
So we made it self-updating.
Hivemind already turns your coding agents' traces into skills. Now that same pipeline maintains your docs.
As your agents learn, your documentation stays current.
Because everything traces back to source, two things get dramatically easier:
• If you get a new team member, they onboard on docs that reflect how things work today.
• Something sticky happens with an agent? See exactly what led to the choices it made.
Get started with Hivemind and stop manually shuffling with docs today.
1
5
413
We talked to hundreds of engineers at AGI Summit this weekend.
We started to notice some patterns around issues.
The same frustrations came up over and over:
• “We are trying to reel in our engineer spend and API cost on tokens”
• “We’re seeing the same bugs come up repeatedly from different engineer output. There doesn’t seem to be any knowledge carry over from things we’ve already fixed across teammates.”
• “Our knowledge base doesn’t seem to be enough to optimize agent output. My agent finally learns our codebase conventions, then the session ends and it's all gone."
Are you facing similar issues internally? What do your agents keep forgetting? 👇
2
3
7
304
start capturing your traces now!
Over the last 90 days, we captured traces from our coding agents to do continual learning.
Here’s what I shared during my keynote on the Continual Learning Loop at the AGI Summit.
We plugged Hivemind into every coding agent at the company: Claude Code, Codex, Cursor, OpenClaw, and Hermes.
- 93% of tokens are used for computer interaction
- 77% of sessions contain at least one correction
- 60% of sessions are judged by an LLM to have been resolved correctly
More interestingly, 1 in 10 sessions leaked a secret credential to an LLM provider.
In-context learning is considered one of the best continual learning methods so far. But how good is it?
Apparently, it is 80% as good as weight updates without all the heavy lifting of fine-tuning.
Naive SFT on coding-agent traces causes the model to collapse. At best, it delivers only marginal improvements. As coding agents use frontier models, distillation can work into smaller models.
But learning from your own LLM traces is very hard. You can easily
- misalign the model
- forget knowledge
- lose capability.
Doing this continually is even harder. You can easily run into cumulative catastrophic forgetting.
Furthermore, learning a new capability from your own traces does not work. So how can we expect the model to develop new capabilities?
How can we dream?
One recipe that worked for us was generating the missing experiences based on failures observed in the traces.
We then used RL environments to train the model and ran it in shadow mode for three weeks. It outperformed Claude Code 60% of the time.
Many people asked, Where should we start?
Start capturing your traces now.
3
222
That's a wrap on AGI Summit 2026
Two days at the Palace of Fine Arts. 500+ of you stopped by the booth and wanted to learn more about the infra layer that levels up your coding agents.
Standard memory tools collect prompts and outputs.
Hivemind turns your coding agents' traces into skills: crystallized once and propagated to every agent on your team.
Legion Code cut token spend 34% (~$12K/month saved) doing exactly this.
Missed us at the booth? DMs are open.
Agents that compound.
4
10
467
Stop guessing what your AI infrastructure costs.
Hivemind is one monthly price with a high usage ceiling. No usage math and no surprise bills.
It pays for itself too:
1. Traces become skills
2. Agents stop repeating work
3. Token spend drops 33%.
Make sure you have continual learning on for your AI agents.
1
1
4
260
We have 3 new updates to Hivemind that are in service of our mission: making organizational agents more cost efficient and effective.
1. Proactive search experience for Claude Code + Cursor. After each user prompt, Hivemind automatically searches for relevant stored traces from the past and passes them to the agent as additional context. This improves reasoning and output.
2. Ability to share skills across teams and devices. Skill sharing is a simpler feature: it syncs the skills a user already has in their system through Deeplake, so everyone on the team can access them. Previously, we were more focused on generating and improving skills. This is about sharing existing skills across the team.
3. Ability to create and assign goals. Any teammate can create and assign goals to another that can persist across sessions until marked completed. Agents can automatically reason when the goal has been reached, and will notify the creator of the goal automatically.
All of these new features work to level up your team, reduce errors and redundant work and lower AI token spend.
Get set up with one command line install.
6
5
9
1,204
I know that feel bro!
The future of the firm is a learning loop in which human capital and token capital compound.
With our new Frontier Co., our ambition is to help every enterprise build its own AI capability, and to help create a frontier ecosystem where every organization can turn its knowledge, workflows, and judgment into its own AI systems that continuously improve. blogs.microsoft.com/blog/202…
1
272
Hivemind crossed 500 stars on github. 🚀
and 2.7K weekly installs.
🌟Star us here github.com/activeloopai/hive…
Hivemind just crossed 250 stars on github
2K weekly downloads on NPM. 🚀
Connect coding agents to a shared brain
> Collect traces into deeplake
> Auto-optimize skills
> Share across agents, machines and teammates
Your agents continuously learn from each other's experience.
Get them to compound your intelligence.
3
7
807
Hivemind just crossed 250 stars on github
2K weekly downloads on NPM. 🚀
Connect coding agents to a shared brain
> Collect traces into deeplake
> Auto-optimize skills
> Share across agents, machines and teammates
Your agents continuously learn from each other's experience.
Get them to compound your intelligence.
Hivemind just crossed 100 stars on github 🌟
github.com/activeloopai/hive…
Beyond memory. Let's compound intelligence!
1
6
11
1,937
Hivemind just crossed 100 stars on github 🌟
github.com/activeloopai/hive…
Beyond memory. Let's compound intelligence!
today, we're going beyond memory.
your org's agents shouldn't just remember what happened. they should learn from their experience.
Hivemind takes agent traces and codifies them into skills every agent on your team can use.
> no more explanations
> no more duplicate work
> no more repeat bugs
we make your intelligence compound.
here's how 👇
1
4
9
1,900
A banger post by our CTO @khustup on how he made Postgres Serverles and spin up under second.
We built a serverless, PostgreSQL-compatible database. Not a modified PostgreSQL deployment.
PostgreSQL provides the interface. DuckDB provides the query execution. Deeplake provides the storage engine.
The architecture makes a different set of tradeoffs than traditional PostgreSQL.
We think those tradeoffs are right for agent workloads: bursty, ephemeral, storage-heavy, and analytical.
Link: deeplake.ai/blog/serverless-…
9
540
The result Free Your Analysts: Stop being "report builders" and become the strategic engine your business needs.
✅ Your analysts become strategists, not data janitors.
✅ Leadership trusts the numbers.
✅ Your GTM strategy is powered by real-time, unified data.
1
1
60
Here’s how it works:
🤖 Automates the integration of disparate sources (CRMs, ERPs, etc.).
🧹 Cleans and manages your messy data taxonomy with AI assistance.
💡 Delivers “just-in-time” intelligence so you can answer strategic questions in seconds.
1
19
Not another dashboarding tool. It's a new way to automate the data harmonization and integration work that consumes your team. This is the bottleneck killing your AI strategy, causing reporting backlogs, and burning out your best analysts.
1
1
42
Your GTM ops team wastes 70% of its time on manual data prep. It’s time to fix it. The endless cycle of manual data preparation, integration, and reconciliation.
We’re introducing Activeloop to unlock AI Data Analysis for GTM Operations.
1
5
10
913
🚀 New free course! Build multimodal #HealthcareAI with Deep Lake. From radiology to drug discovery, integrate text, images & more with LLMs + AI search.
1
4
12
505
Launched Activeloop L0 on @ycombinator. L0 turns multimodal documents into cited answers with state-of-the-art accuracy. Check out the link below!
4
12
42
12,250
Introducing Deep Lake AI Knowledge Agent. Conduct Deep Research on your data, no matter its modality, location, or size.
Come celebrate with us @ProductHunt!
producthunt.com/posts/deep-l…
10
22
1,779
A week ago, we've launched our Ai Knowledge Agent.
Today, we are excited to announce that we've been recognized as an Emerging Specialist in Generative AI Engineering and AI Knowledge Management Apps/General Productivity by @Gartner_inc.
Even further, with the introduction of our AI Knowledge Agent that performs Deep Research on multi-modal private data, Gartner has placed us the closest to an Emerging Challenger compared to our peers in our quadrant.
We consider our positioning in the Emerging Specialist
quadrant by Gartner as confirmation of our mission to help answer tough questions on multi-modal data in highly complex cases - in drug discovery, financial analysis, medical claims processing, patent search, and beyond.
To celebrate the inclusion - we're launching on @ProductHunt with an introductory offer! Click subscribe to be notified of the release and get exclusive, free access to Deep Lake Knowledge Agent! (link in next tweet)
ALT ai knowledge management quadrant
ALT ai engineering quadrant
1
7
13
768
⚡ Speed & Scalability:
Our “index-on-the-lake” tech enables sub-second queries even over 2B+ records—no expensive in-memory caching required (we're 10x cheaper vs in-memory DBs). Fast, cost-effective AI search that scales with your data needs.
ALT architecture diagram
1
2
46
🔍 How It Works:
1️⃣ Connect & index your data
2️⃣ Ask a natural language question
3️⃣ Our agent translates it into targeted sub-queries across modalities
Result? A comprehensive answer backed by text, images, and even data figures!
1
2
58
Deep Lake isn’t just for text! It processes images, video, audio, & structured metadata with help of VLMs (Vision Language Models). For example, research teams can synthesize insights across patient charts, lab tests, and MRIs in one query.
ALT multimodal ai search on medical data
1
2
91
Unlike other AI assistants that search only the web, Deep Lake integrates both public & private data—deployable on your S3 or Azure. Whether it’s internal reports, research, or proprietary IP, your data stays secure & compliant. Enjoy fine-grained access control and SOC2 Type II compliant deployment!
1
2
108
Why did we build this? Enterprises lose 21–25% productivity (up to $20K/employee/year) on manual searches. Imagine paying your team to play hide-and-seek with your own data! Deep Lake fixes that.
ALT connect any data source
1
2
66
Summarize facts across multiple tables and papers, with precise references.
Try now via chat.activeloop.ai
1
2
81
The below is taken from Marcel Proust's novel In Search of Lost Time (French: À la recherche du temps perdu) - one of the longest books ever written (at more than 1150 pages in PDF format).
Try now via chat.activeloop.ai
1
2
107
Key findings from a 200+ page paper, gleaned both from text and the corresponding figure.
Read more on: activeloop.ai/resources/intr…
1
1
3
152
Today, we're introducing Deep Research on Your Own Multi-Modal Data.
Just connect any source from your cloud or local storage and ask any question for a thoughtful response formulated by our Knowledge Agent.
Example answers before we dive in:
2
6
26
3,175
ColPali made it possible to use VLMs for document retrieval instead of manually feature engineered OCR pipelines with equal quality on raw text and much better on in-document figures, and tables. But it comes at a cost.
The cost is storage scalability:
ColPali’s embeddings require 256 KB per page, a 30x larger memory footprint compared to a single embedding.
Also, ColPali relies on a multi-vector retrieval based on ColBERT’s late interaction mechanism. It is not natively supported by many vector retrieval frameworks or databases, which further increasing the engineering complexity for deployment.
That is, unless you use Deep Lake by Activeloop, which companies like @Bayer, Flagship Pioneering, or @Matterport do.
What can Deep Lake do?
1. Offload any amount of multi-modal data on your S3/GCP/Azure or local file system, instead of relying on costly in-memory storage.
2. Apply ColPali magic for contextualized data embedding that encodes the entire page’s structure, text, and visuals.
3. Use MaxSim operator for maximum similarity scores across tokens or patches - no extensive engineering is needed as it is natively supported.
4. Enjoy sub-second queries across billions of rows of multi-modal data types like embeddings, images, and text. Capture all the nuances of your data (e.g., data from text mentioned in a figure or an image inside of the PDF).
In this practical guide with 1000s of user manuals for printers, you can ask questions like ’’My printer is not out of ink but shows error code E01—how do I clear it?“, “How to clear a jam in Canon Pixma?” etc.
ColPali is revolutionary, as it was first time to replace OCR and will quickly get much better at capturing document context as VLMs improve.
Scaling it on all your data to truly leverage your organization’s multi-modal knowledge AND not breaking a bank is hard - unless you use Activeloop.
Full guide : activeloop.ai/resources/col-…
6
12
1,018
We've Launched Deep Lake 4.0 at RetrieveX on October 17th, with 200+ AI Data Leaders from companies like Bayer, Tesla, Ally Financial, LinkedIn, Adobe, Walmart, Netflix, DoorDash, RingCentral, Zendesk and more attending. It was a blast!
AI data retrieval systems today face 3 challenges: limited modalities, inaccuracy, and high costs at scale. Deep Lake 4.0 fixes this via true multi-modality, higher accuracy, and slashing query costs by 2-10x with our unique index-on-the-lake technology.
But also, the conference was packed with amazing talks from:
Armen Aghajanyan from @AIatMeta / FAIR presenting his learnings from training Meta Chameleon and his thoughts on the future of multi-modal LLMs and RAG.
Rob Ferguson, Head of AI, Microsoft for Startups presenting who and why leads in retrieval augmentation.
@aronchick , CEO @ExpansoIO and Co-Creator of Kubeflow and Bacalhau presenting on executing one's models wherever your data and users are.
Steffen Vogler, @Bayer, presenting the next-gen approach to AI in Healthcare.
Ian Trase, @FlagshipPioneer, presenting how the company's Flagship Pioneering arm makes big leaps in scientific research with GenAI.
@jiayq, Founder @LeptonAI and co-creator of @PyTorch and CAFFE, presenting on building enterprise-ready LLMs.
@sdianahu, Group Partner at @ycombinator, moderating a panel on scaling efficiently in the age of GenAI with HerculesAI CTO Gevorg Karapetyan and @omneky CEO Hikari Senju.
Sazzadur Rahman, @spotterstudio, presenting optimizing content creation on YouTube for contextual search with up to billion-scale search.
@viglovikov , CEO and Creator of @albumentations , presenting efficient data augmentation in vision AI during his practical workshop.
Vivek Gangasani, @awscloud, presenting how to use Bedrock for GenAI and combine it with Activeloop Deep Lake to improve Retrieval
Kelly Peng, Kura founder, presenting on building multi-modal AI that sees, remembers, generates.
Bill Sun, Generative Alpha, presenting AI trader with reasoning capabilities powered with GenAI.
@activeloop with @khustup and @MikayelHarut presenting sub-second, multi-modal AI search on object storage.
Major thanks to our sponsors at @awscloud, @intel, and @Microsoft for supporting our vision and the event!
2
13
28
2,645
We've unveiled Deep Lake 4.0 at #RetrieveX on Oct 17 to 200+ AI Data Leaders. It is already in production with @Bayer, @FlagshipPioneer, @Matterport, and @SpotterCreators. It's 10x more cost efficient than leading in-memory databases. Why did we release it?
AI data retrieval faces five core challenges: limited modalities, accuracy issues, high costs to build it right, limited memory, and manual workflows. Deep Lake 4.0 addresses all these with the following new features:
Index-on-the-Lake: Enables sub-second queries directly from any object storage and cross-cloud.
10x Cost-Efficiency: Index-on-the-lake eliminates the need for costly in-memory storage and large clusters. Deep Lake provides rapid, scalable search without the overhead.
High Accuracy: Utilize multiple indexes (embedding with quantization, lexical, inverted, etc.) for rapid search on object storage with minimal caching, ready for neural search technologies like ColPali.
True Multi-Modality: Supports diverse data types with enriched metadata.
Enhanced Performance: Up to 10x faster reads/writes due to migrating low-level code to C++.
Learn more about Deep Lake 4.0 in our release post.
activeloop.ai/resources/deep…
4
10
567
RetrieveX is on!
🔥 @ArmenAgha from @Meta just kicked us off with his keynote on why retrieval is needed from first principles, highlighting state-of-the-art uses of retrieval beyond traditional settings, specifically in multimodal applications.
🔥 Followed by @DBuniatyan, CEO of @activeloop, who discussed AI search on Data Lakes and how enterprises can organize complex unstructured data.
2
8
301
We’re ready for RetrieveX today, are you?
👉retrievex.co/
Final chance to grab your tickets at 75% off with promo code FINAL for the last tickets left. See you there!
1
130
When it comes to healthcare AI, mistakes matter.
Whether it's real-time surgical assistant error, bias in MRI scans, or incorrect information taken as input for drug discovery, mistakes can cost a lot of wasted budget on a wrong compound, or even lives.
(If you're a Data Leader in AI - apply today via this link and we may grant you a complimentary ticket! retrievex.co/application)
@Bayer and Steffen Vogler's team, as well as Ian Trase's team at @Flagship_PI / @FlagshipPioneer are addressing these challenges together with Activeloop by focusing on retrieval for AI. Both are presenting at RetrieveX next week!
Join RetrieveX, the Conference for AI Data Leaders on October 17 to learn more.
#LLMs #RAG #AI
7
10
425
@Techweek_ is here! Grab the last remaining spots for our meetup with @llama_index - excited to speak alongside @jerryjliu0 on Multi-Agentic Workflows in Production.
Join to learn how we use AI Agents for radically easier data ingestion, exploration, and more!
RSVP here: lnkd.in/gKnSRAy3
Major thanks to #TechWeek for having us!
5
4
56
RetrieveX is coming in 2 weeks on Oct 17 in SF - and it will be 🔥
Join the creators of Meta LLama, PyTorch, CAFFE, Kubeflow, and leaders from Microsoft, AWS, Bayer, YCombinator at the best Retrieval for AI Conference.
Get a 2 for 1 deal until Oct 7!
retrievex.eventbrite.com
1
11
16
1,052
Tickets on sale via the website - Take extra 25% OFF by getting your ticket today!
eventbrite.com/e/retrievex-t…
1
4
116
Join RetrieveX - THE conference on the Future of Retrieval for Generative AI.
When: Oct 17
Where: In-person, San Francisco, CA
Capacity: 300 Executives. Ticket prices going up on Sep 9!
Executive in GenAI? We've curated this event with our customers and partners exclusively for leaders building high-accuracy, multimodal workflows. Come hear from innovators from @Microsoft, @Bayer Radiology, @YCombinator, @FlagshipPioneering, @LeptonAI, @Expanso, @Omneky, @Cresta, and more.
Build a solid data foundation for your GenAI. Act fast—tickets are limited!
retrievex.co
3
5
16
1,192
We've teamed up with top AI minds, #IntelDisruptor, and @towards_AI, to craft the Impossible GenAI Test. Only 1 in 20 can succeed. Think it can be you?
- 30 Questions, 6 Topics, 40 Minutes
- Questions do not repeat, and vary in difficulty
- Wrong answers are penalized
4
5
14
1,430
LazyLLM is a dynamic token pruning method for efficient long context LLM inference, accelerating the pre-filling stage of LLama 2 7B by 2.34x while maintaining accuracy.
ALT How LazyLLM differs from standard LLM
1
3
109
