25 Photos and videos
The Manus AI move from Mark Zuckerberg is interesting, but it also exposes a strange gap in Meta’s AI strategy.
Meta wrote one of the biggest success stories in open AI with Llama. Half the open model ecosystem fine-tuned or built on top of it.
Yet while the industry was shipping products, agents, and platforms, Meta itself shipped almost none.
No major AI platform.
No breakout developer tool.
Now the Manus move feels less like innovation and more like catching up.
How did the company that powered the open model ecosystem fall this far behind in products?
x.com/beginbot/status/203145…
1
2
32
Gider retweeted
Mar 10
Mark Zuckerberg acquiring AI companies
268
551
10,084
1,375,654
Claude Opus 4.6 partnered with Mozilla and found 22 CVEs in Firefox in just 2 weeks — 14 high-severity. 🔥
That's nearly 1/5 of ALL high-severity bugs Mozilla fixed in 2025. First one spotted in ~20 minutes. Fixes shipped in Firefox 148.
AI security scanning is accelerating fast.
But one model can still miss things due to bias or blind spots...
In Dvina Code we run parallel agents across GPT-5.4 Claude Opus 4.6 others at the same time → real-time comparison, auto-fix loops, testing, and reports. All privacy-first.
Single-model era ending? Which setup wins for you in security work? 👇
@dvinaai
dvina.ai - Dvina Code
#AICybersecurity #ClaudeOpus #GPT54 #MultiModelAI #Firefox
1
4
295
Big news: GPT-5.4 is now live as the default model in Dvina Code, available to all users starting today 🔥
What’s new?
• 1M token context (experimental): Work across massive codebases without losing the thread
• Native agentic computer-use: True parallel agent swarms that execute and complete complex tasks
• Improved reasoning token efficiency: Smoother performance for advanced developer workflows
• Privacy unchanged: Your code is never used for training
This upgrade genuinely changes the feel of building with AI. Agents aren’t just assisting anymore, they’re collaborating.
To celebrate the launch, we’re still running our 1-month Dvina Plus FREE offer. Perfect time to explore frontier-level tooling.
#GPT54 #DvinaCode #Coding
Just dropped in Dvina Code: GPT-5.4 🔥
More powerful coding/reasoning, native agentic features, 1M token context and huge efficiency gains.
Grab the latest update and feel the difference!
1
155
We recently had an infrastructure overload, so Dvina Code was temporarily unavailable in some regions. The issue is now fixed, and access should return soon depending on your region as the changes propagate.
Thanks for the strong interest and patience. The demand makes us realize we’re building something dev community truly want.
3
9
815
Linear's agents directory has become incredibly powerful — seeing seamless delegation to tools like Cursor, Codex, and others is transforming workflows.
Dvina Code recently launched as a strong addition in this space: it runs parallel frontier-model agents (e.g., Claude Opus 4.6 GPT-5.3 Codex, Kimi K 2.5) in a privacy-focused desktop GUI environment, enabling collaborative real-time coding without CLI friction.
It would be a valuable enhancement to include @dvinaai in the agents directory as well. Delegating a Linear issue to the swarm for parallel planning, implementation, vulnerability fixes, and automated PRs could significantly boost productivity for many teams.
Would Linear consider adding it? The combination feels like a natural fit.
Launch post with details below 👇
@cjc @linear
#coding #ai #linear
We built a GUI-first agentic coding platform with Claude Opus 4.6 on the free plan, giving away 1 month Plus for launch
Hey Everyone,
We've been quietly building Dvina Code and are doing a proper launch today. Wanted to share it here since this community cares deeply about agentic coding tools.
The core problem we're solving: Most agentic coding tools are either CLI-heavy, cut you off mid-task with usage limits, or lock the best models behind expensive tiers. We wanted to fix all three.
What Dvina Code does differently:
Frontier models with real headroom: Claude Opus 4.6, Opus 4.5, Sonnet 4.5, GPT-5.3 Codex, Kimi K 2.5 and more.
Parallel sessions: run multiple coding agents simultaneously. One writing a feature, one fixing a bug, without losing context on either.
Custom agents: define instructions, pick a model, connect tools, scope to a repo. Build a reviewer agent, a test agent, a security agent, a triage agent, once, reused across the team.
GUI-first design: built for people who want a smooth desktop experience, not another terminal wrapper.
Privacy: your sessions are never used to train any model, including ours.
Available on macOS (Apple Silicon Intel) and Windows.
🎁 Launch offer: 1 month free Plus membership for early users.
Download details: dvina.ai/p/dvinacode/dvinaco…
Happy to answer any questions. Would genuinely love feedback from people who push these tools hard.
#coding #ai
1
3
450
Coding with AI wasn’t supposed to feel this fragmented.
Models have become significantly more powerful.
Yet the workflow often feels heavier.
Too many panels.
Too much context switching.
Too little sustained flow.
This is the gap we kept coming back to.
AI shouldn’t replace the developer.
It should extend their capacity.
It should feel like:
– You remain fully in control
– The agent understands your codebase and intent
– Work progresses in parallel
– Focus lasts longer than a few minutes
That belief led us to build Dvina Code.
A calmer interface.
Parallel agents with real context awareness.
The first step toward our “live context” vision.
We’re now opening early access.
1 month of Plus, complimentary.
Structured limits, meaningful headroom. Enough to build something real.
If you’re seriously building with AI today, I’d value your perspective.
What still feels broken?
What’s missing?
What would justify switching tools?
Comment “DVINA” or DM @dvinaai.
#AIDev #DevTools #Coding
We built a GUI-first agentic coding platform with Claude Opus 4.6 on the free plan, giving away 1 month Plus for launch
Hey Everyone,
We've been quietly building Dvina Code and are doing a proper launch today. Wanted to share it here since this community cares deeply about agentic coding tools.
The core problem we're solving: Most agentic coding tools are either CLI-heavy, cut you off mid-task with usage limits, or lock the best models behind expensive tiers. We wanted to fix all three.
What Dvina Code does differently:
Frontier models with real headroom: Claude Opus 4.6, Opus 4.5, Sonnet 4.5, GPT-5.3 Codex, Kimi K 2.5 and more.
Parallel sessions: run multiple coding agents simultaneously. One writing a feature, one fixing a bug, without losing context on either.
Custom agents: define instructions, pick a model, connect tools, scope to a repo. Build a reviewer agent, a test agent, a security agent, a triage agent, once, reused across the team.
GUI-first design: built for people who want a smooth desktop experience, not another terminal wrapper.
Privacy: your sessions are never used to train any model, including ours.
Available on macOS (Apple Silicon Intel) and Windows.
🎁 Launch offer: 1 month free Plus membership for early users.
Download details: dvina.ai/p/dvinacode/dvinaco…
Happy to answer any questions. Would genuinely love feedback from people who push these tools hard.
#coding #ai
1
130
We built a GUI-first agentic coding platform with Claude Opus 4.6 on the free plan, giving away 1 month Plus for launch
Hey Everyone,
We've been quietly building Dvina Code and are doing a proper launch today. Wanted to share it here since this community cares deeply about agentic coding tools.
The core problem we're solving: Most agentic coding tools are either CLI-heavy, cut you off mid-task with usage limits, or lock the best models behind expensive tiers. We wanted to fix all three.
What Dvina Code does differently:
Frontier models with real headroom: Claude Opus 4.6, Opus 4.5, Sonnet 4.5, GPT-5.3 Codex, Kimi K 2.5 and more.
Parallel sessions: run multiple coding agents simultaneously. One writing a feature, one fixing a bug, without losing context on either.
Custom agents: define instructions, pick a model, connect tools, scope to a repo. Build a reviewer agent, a test agent, a security agent, a triage agent, once, reused across the team.
GUI-first design: built for people who want a smooth desktop experience, not another terminal wrapper.
Privacy: your sessions are never used to train any model, including ours.
Available on macOS (Apple Silicon Intel) and Windows.
🎁 Launch offer: 1 month free Plus membership for early users.
Download details: dvina.ai/p/dvinacode/dvinaco…
Happy to answer any questions. Would genuinely love feedback from people who push these tools hard.
#coding #ai
13
27
165
179,344
NVIDIA’s moat was general-purpose GPUs.
But AI compute is already defecting:
@cerebras escapes GPU scale limits
@GroqInc removes runtime overhead
@taalas_inc eliminates the GPU entirely
If models become hardware,
the CUDA era was just a phase.
@nvidia tried to buy Groq.
It may need to move faster on Taalas.
#taalas #groq #nvidia
54
Gider retweeted
Feb 21
We started using Dvina Code in the office 2 weeks ago.
Result?
Engineers delivered 7x more bug fixes, features, and improvements across parallel projects including our main AI platform Dvina.
Next week we’re launching Dvina Code for everyone.
• Free Opus 4.6
• Higher usage limits
• Cheaper membership options
For launch, I’ll share a 1 month Plus promo code.
Comment if you want it 👇
#vibecoding #claudecode #buildinpublic #indiehackers
3
2
4
370
Gider retweeted
Building a more reliable web together is what it’s all about.
1
1
118
Gider retweeted
Feb 14
Next week, something new wakes up. Dvina Code is almost here.
Not just another AI coding tool… a step toward true agentic development with live context awareness.
Think bigger sessions.
Think smarter workflows.
Think Opus 4.6 with cheaper pricing and higher usage limits (even on the free tier).
If you’ve been waiting for AI that actually codes with you instead of just answering…
Don’t miss the launch. 👀 @dvinaai
1
1
6
3,734
Launched on @ProductHunt : Dvina for your Inbox! 🤖📧
Private AI manages Gmail/Outlook, connects 120 apps live data → zero chaos, real insights.
Launch deal: Plus package 3 months FREE w/ premium models!
Upvote & try →
producthunt.com/products/dvi…
#PrivateAI #ProductHunt #AI
3
2
6
523
After months of late nights and endless iterations, super excited to finally launch Dvina for your Inbox on @ProductHunt today! 🚀
Connect your Gmail & Outlook, bring in 120 apps live databases, and let private AI actually manage your inbox turn scattered data into real-time insights.
Would mean the world if you check it out, upvote, or drop a comment with your thoughts ❤️
producthunt.com/products/dvi…
#buildinpublic #AI #ProductHunt"
2
3
47
Just launched: Dvina for your Inbox on @ProductHunt! 📧✨
The world's most connected private AI now manages your Gmail & Outlook inbox while unifying 120 apps, live databases, docs & real-time data in one cognitive space.
No more scattered info - analyze, create & decide faster.
Upvote & try it if this resonates! 👇 producthunt.com/products/dvi…
#AI #InboxZero #Productivity"
1
38
BitNet / bitnet.cpp is seriously cool 🔥
1.58-bit LLMs screaming on CPUs - up to 6x faster inference, 80% less power. Real milestone for local/edge AI.
But GPUs aren't dead 😅
They still own training & high-throughput. This just opens a killer budget/low-power lane for everyone else.
AI hardware family just got bigger - CPUs join the party.
Rich stay on GPU clusters, mortals run 70B models on laptops.
Tried bitnet.cpp yet? Drop your tokens/sec hardware below 👇
CPU rising in 2026 or GPUs forever dominate?
Jan 29
Microsoft killed the GPU mafia 🤯
They finally open-sourced their 1-bit LLM inference framework called bitnet.cpp. It lets you run 100B parameter models on your local CPU without GPUs.
- 6.17x faster inference
- 82.2% less energy on CPUs
100% Open Source.
Community note
This work and repository are more than 1 year old - see the GitHub and attached paper.
github.com/microsoft/BitN…
98
Gider retweeted
Jan 19
A Reddit user claims to have scanned approximately 20,000 Vibe-coded indie apps that utilize Supabase, and he found that 1 in every 9 of them has their database access key publicly available.
reddit.com/r/Supabase/s/23yI…
1
4
86