Drokski

Surfing.

[Dot @usedotai]

Gemini 3.6 Flash is now live and fully anonymous on Dot.

Available across DotChat and DotCode, and 10-13% cheaper than our largest competitors.

Enjoy the cheapest inference on the market with Dot: https://t.co/3UiVElkGAV

[stage @stagedhappen]

The Dot ecosystem just got even more private. Introducing Dot Encrypted Sync.

A new local-first architecture that encrypts your conversations inside your browser before they ever leave your device.

Dot’s sync service stores ciphertext only. It never receives readable messages, conversation titles, or history.

Key capabilities:
-Client-side AES-256-GCM encryption
-Independent encryption and authentication credentials
-Automatic multi-device synchronization
-Offline retry and conflict resolution
-One-click sync pause and permanent server deletion
-Local-first by default—sync is always optional

There are no cloud accounts to manage and no additional recovery keys.

Your existing encrypted Dot recovery file restores your credits, subscription, API access, and encrypted conversations across devices.

Your synchronized conversation history belongs to you, not your AI provider.

Protect everything you built, architecturally, with @usedotai.

Forward.

[stage @stagedhappen]

We have completed our largest platform overhaul since Dot launched, fundamentally improving both the user experience and the capabilities of the platform.

We have:
-Rebuilt our entire subscription architecture, transforming Pro from a credit bundle into a true subscription experience.

-Completely redesigned Chat, Image and Video around usage-based access, removing frontier model paywalls and unlocking every available model under a single subscription.

-Introduced live usage tracking with rolling 5-hour and weekly allowance windows, alongside a new account dashboard showing plan status, usage and private credits in real time.

-Redesigned navigation, model discovery, subscriptions, checkout, account management, conversation history and recovery flows across the platform.

-Rebuilt the backend usage engine with server-side accounting, reserved inference allocation, signed subscription entitlements and significantly improved reliability.

-Expanded wallet recovery to support both standard wallets and Base smart-contract wallets, while allowing complete restoration of subscriptions, API keys and private credits.

Additionally, private credits are now fully separated from subscriptions. They never expire and remain dedicated to DotCode, API access and optional pay-as-you-go inference, while Chat, Image and Video are now powered directly through your active subscription.

A faster platform. A cleaner experience. A significantly stronger foundation for everything we're building next.

Let the week begin.

Dot.

[stage @stagedhappen]

One prompt. Kimi K3. $0.0013.

Kimi K3 is now live on DotCode, and producing stunning websites with one, singular, prompt.

The best part? Dot offers the cheapest inference on the market, producing practically zero coding cost.

Try it now at: https://t.co/QGe5SFr9jB

[stage @stagedhappen]

We spent $40,000 on infrastructure, and have now trained our own model.

Dot Loom Conductor v0.1 is live: our first trained orchestration model for multi-modular AI systems.

DLC is a 14B control model that decides how much inference a request receives, which models participate, and what each model is allowed to see.

Training our first model fundamentally changes what Dot is capable of becoming. Instead of relying exclusively on frontier providers, we can now build intelligence purpose-designed for our own infrastructure.

Every future optimisation, routing decision, orchestration policy, and verification strategy can itself become learned, creating a system that improves with every deployment rather than remaining statically engineered.

[Drokski @Drokskichain]

@stagedhappen Incredible spike !

[stage @stagedhappen]

Dot Loom, our intelligent inference orchestration framework, has now surpassed 164 GitHub stars following its architectural overhaul.

Dot Loom's adoption is driven by its ability to address a fundamental limitation of modern AI systems: no single model consistently outperforms across every task.

Loom treats frontier models as composable infrastructure rather than isolated endpoints.

Builders can orchestrate OpenAI, Claude, DeepSeek, Qwen, Dot, Ollama, or any OpenAI-compatible provider within a single execution graph, assigning specialized models to drafting, reasoning, verification, or refinement.

The orchestration layer is designed to remain computationally efficient:
-Lean → maximum 1 inference call
-Balanced → maximum 2 inference calls
-Strict → maximum 3 inference calls
-No dedicated routing model required
-Configurable latency, credit, and call budgets
-Full execution traces, raw model outputs, verifier reasoning, receipts, and reproducible benchmarks

Equally important, we published the complete benchmark methodology (including failure cases, trade-offs, and performance regressions), not just the strongest results.

Transparent evaluation matters as much as orchestration itself.

This initial traction is an encouraging signal that developers increasingly value provider-agnostic inference, verifier-backed generation, and reproducible AI systems over opaque single-model abstractions.

Check out our Dot Loom at: https://t.co/W5w3xo7bLz

[stage @stagedhappen]

Dot has processed 477M tokens over the last 9 days. Yesterday, our platform consumed 63M tokens, making it our third-largest usage day since launch.

Alongside this growth, Dot has now accrued $33,700+ in total inference commitments across the ecosystem.

Current breakdown:
-$13.2k+ in inference credits
-$4.2k+ in native $DOT inference spend
-$16.3k enterprise agreement

DotCode subscriptions continue to scale alongside platform usage.

Current Monthly Recurring Revenue:
Pro:
-18 paid in $DOT ($216)
-14 paid in USDC ($210)
→ 32 users | $426/month

ProX:
-9 paid in $DOT ($567)
-6 paid in USDC ($474)
→ 15 users | $1,041/month

Max:
-5 paid in $DOT ($595)
-2 paid in USDC ($298)
→ 7 users | $893/month

Current totals:
-32 subscribers paying through the native $DOT rail → $1,378/month
-22 subscribers paying in USDC → $982/month

Total Monthly Recurring Revenue: $2,360

Our objective remains unchanged: continue improving the efficiency of the Dot ecosystem, grow our user base via new marketing funnels, relentlessly prioritize retention, and keep building products people genuinely use.

Forward.

[stage @stagedhappen]

At Dot, we understand deeply that the future of AI does not rest in the hands of one singular model, but rather, intelligent orchestration.

That is why we open-sourced Dot Loom, and today, we are shipping its biggest upgrade yet.

We gave OpenAI, Claude and Dot the same 6 backend nightmares:
-Billing races
-Tenant data leaks
-Webhook replays
-Streaming refund bugs
-OAuth failures
-SSRF attacks

Results from a blinded Deepseek judge:
OpenAI: 100% quality, 12.67 credits
Claude: 88.3% quality, 3.50 credits
Dot: 70% quality, 1.00 credit

Claude reached 88.3% of the top score using 28% of the credits. Dot reached 70% using 8%.

But Dot Loom is not an OpenAI or Claude wrapper, those are just recognizable examples.

You can combine any models you want:
-OpenAI writes, Claude reviews.
-Claude writes, DeepSeek reviews.
-Qwen drafts, OpenAI finalizes.

A local Ollama model handles easy work, then a stronger cloud model checks risky requests.

Loom is model and provider agnostic. It works with Dot, OpenAI-compatible APIs, @OpenRouter, @Ollama, LM Studio and local models.

With the Dot API, one key and one endpoint gives you access to @OpenAI, @ClaudeAI, @Deepseek_ai , @Alibaba_Qwen and Dot models. You can also bring your own providers and configure every role yourself.

Loom controls how much inference each task receives:
• Lean: 1 model call
• Balanced: writer + reviewer
• Strict: writer + critic + finalizer
• No paid router call
• Model-specific credit, call and latency budgets
• Raw answers, costs, token usage, judge reasons and receipts

Our goal is to use cheaper models where they're enough, stronger models where they matter, and independent models when verification is critical.

Open source, with raw benchmark results, charts, receipts, and reproducible runners: https://t.co/W5w3xo6DW1

[Drokski @Drokskichain]

@stagedhappen

[stage @stagedhappen]

We’ve rebuilt the entire image experience on Dot.

-one calm prompt bar → submit → a true-ratio canvas materializes
-a living mark + red trace circles the frame while your image forms
-blur-to-sharp reveal when it lands. no fake progress bars, ever
-real elapsed time, model + cost always visible: 1 cr / image on lust-v7
-rebuilt for mobile: canvas-first, swipeable prompts, crisp everywhere

Zero account. Zero prompts stored. Your gallery lives only in your browser.

generate in full privacy → https://t.co/QIRgoLbJdy

[stage @stagedhappen]

The @BytePlusGlobal Seedream V5 Pro is now live on Dot, and priced 10% cheaper than our competitors.

Built for high-end product imagery, cinematic scenes, strong composition and precise prompt adherence, with native 2K generation across multiple aspect ratios.

At 10 credits per image ($0.10), DotImage is approximately 10% cheaper than the largest provider offering the same model.

Zero prompt history retention.
No generated images stored by us

Available now at: https://t.co/QIRgoLbJdy

[stage @stagedhappen]

Our first Dot-24B training run is now underway.

Current configuration:
-QLoRA fine-tuning on a 24B base model
-NF4 4-bit quantization + BF16 compute
-92M trainable parameters (0.39%)
-Single NVIDIA H200
-Batch size 2 × gradient accumulation 16
-3 epochs over a 12.8K seed dataset
-45s/iteration (~15h wall clock)

Current training loss is converging as expected.

This run isn't intended to produce the final model, it's validating the training pipeline, memory profile, optimizer stability, checkpointing, and data flow before scaling.

The next phase focuses on synthetic dataset generation and large-scale instruction tuning, with evaluation suites determining final alignment, reasoning performance, and refusal behavior.

Weights and evaluation results will be published shortly.

Forward.

[stage @stagedhappen]

We just made the Dot ecosystem even cheaper, without sacrificing user privacy.

DotChat now supports privacy-conscious prompt caching across 21 models, spanning GPT, Claude, Kimi, Qwen, GLM, DeepSeek and Nemotron.

Instead of repeatedly billing the same conversation history at the full input-token rate, supported models can reuse warm context at their lower cache-read price.

We ran a live full-model matrix using identical long-context prompts:
- Cold requests: 2.93s average
- Warm cached requests: 2.22s average
- Average latency reduction: 24.4%
- Median latency: 2.11s → 1.79s
- 18 models returned confirmed cache hits

Caching is activated through opaque, conversation-scoped keys. Cache entries are never shared between users or conversations, and Dot does not maintain a plaintext prompt-cache database.

The backend separately meters:
- Uncached input tokens
- Cache-write tokens
- Cache-read tokens
- Output tokens

This means repeated context can be processed faster and billed at the lower cache-read rate when supported. Unsupported or temporarily unavailable caching automatically falls back to normal inference.

Privacy will never become a trade-off for performance. It has always mattered to Dot, and it always will.

*Live results vary with provider load, output length and model availability.*

Try a cheaper Dot at: https://t.co/QIRgoLbJdy

[Drokski @Drokskichain]

@stagedhappen 55k$ total worth of infrastructure, a working coding agent , a working private chat bot , MCPs , APIs , guys what do you want more ? base:0x23a2847d772803f9efc64b4277b782b06296fe51

[stage @stagedhappen]

Over the past 2 weeks, we’ve committed $40,000 toward physical H200 server infrastructure to reduce dependency on external inference providers and bring more of Dot’s AI stack under our own control.

The main objective? To fine-tune our own Dot edition of Mistral Small 24B from @MistralAI.

We brought up an NVIDIA H200 SXM5 node with 141GB HBM3e and ran a complete local fine-tuning pipeline on Mistral Small 24B open weights.

The first run:
→ NF4 QLoRA fine-tuning → 92.4M trainable parameters (0.39% of the base model)
→ 50 training steps completed in 221 seconds
→ 91-100% sustained GPU utilization
→ Reproducible 370MB Dot adapter checkpoint produced

Next, we begin building toward Dot Mistral 24B: a Dot-trained, Dot-hosted model built from open weights, with measurable behavior and a privacy-first deployment path.

By fine-tuning open-weight models ourselves, Dot can build specialized intelligence around our privacy architecture, our users, and our ecosystem, without relying entirely on external providers.

This creates a path toward Dot-owned models:
→ Dot-trained
→ Dot-hosted
→ Privacy-first
→ Built for real-world agentic workloads

Dot.

[Dot @usedotai]

We're rapidly approaching Claude Code level outputs on DotCode, whilst integrating multiple models, and exposing cheaper routing.

Check out a website we just made for $0.50, with one, singular prompt: https://t.co/NP6mxqakrB

[stage @stagedhappen]

Our subscription packages have been live now for under 24 hours, and we've already hit a total monthly recurring revenue of $1,191.

Breakdown:
Pro:
-11 paid in $DOT ($132)
-8 paid in USDC ($120)
→ 19 users | $252/month
ProX:
-5 paid in $DOT ($315)
-3 paid in USDC ($237)
→ 8 users | $552/month
Max:
-2 paid in $DOT ($238)
-1 paid in USDC ($149)
→ 3 users | $387/month

Current totals
-18 subscribers paying through the native $DOT rail → $685/month
-12 subscribers paying in USDC → $506/month

Our focus now is simple: improve the efficiency of the Dot ecosystem whilst growing our userbase, relentlessly prioritizing user retention.

https://t.co/QIRgoLbJdy

[stage @stagedhappen]

An update on Dot:

On June 19th, we pushed Dot into monetized access, a monumental day for us. At no point on that day, did we ever believe we would reach the numbers I am about to present to you all.

Over the last 3 weeks, we’ve seen rapid adoption across users, developers, and enterprise customers:
-$8.5K+ in inference credits purchased
-$3.1K+ in native $DOT inference spend
-$16.3K enterprise agreement signed
In just 21 days, Dot has generated $27.9K+ in monetized activity.

At current pace, this represents an annualized revenue run rate of +/- $485K.

And our usage continues to accelerate:
→ 287M+ tokens processed since monetization
→ 397M+ tokens processed since inception

Soon, we will be onboarding an infrastructure engineer to help us prepare for the next phase of Dot.

We will be allocating $40,000 toward physical servers, pushing Dot closer to the hardware layer and further strengthening our privacy-first infrastructure.

More on this soon.

[stage @stagedhappen]

The whole GPT-5.6 family of models are now available in DotChat. Fully private, and 10-20% cheaper than our largest competitors.

Access them today at: https://t.co/QGe5SFr9jB