Daft.Exe

Daft.Exe is a full-stack interactive web app OS with some daft features like a ZK Privacy Bunker, AI Agent Swarm Command Center and an AGI Dream Lab. Come in.

[Daft.Exe @DaftExeOS]

The time is nearly upon us...

$DAFT

[Daft.Exe @DaftExeOS]

Back to $DAFT

[Daft.Exe @DaftExeOS]

@VillianOfCode 's previous work are projects that you have all probably heard about on-chain. All have reached high valuations 🤫

[Villano @VillianOfCode]

GM World.

Our Q&A has been rescheduled for today, we will be going through the latest updates and the new token gating details!

We will update with an exact time soon.

Meanwhile you can read: https://t.co/QcrYLsW1Rw

CA: F2iqrq41NfJoZnK68UBmrbZrLPkPDaJzZHyZFbDdpump

[Villano @VillianOfCode]

GM World.

Our Q&A has been rescheduled for today, we will be going through the latest updates and the new token gating details!

We will update with an exact time soon.

Meanwhile you can read: https://t.co/QcrYLsW1Rw

CA: F2iqrq41NfJoZnK68UBmrbZrLPkPDaJzZHyZFbDdpump

[Daft.Exe @DaftExeOS]

SABLE is the part of Nous’ Angels that makes the whole thing feel serious. Anyone can make agents generate findings. The hard part is making sure those findings are grounded, traceable, challenged, and safe before they become “truth.” SABLE acts like the adversarial reviewer.

[Villano @VillianOfCode]

The SABLE governance layer might become the most important part of Nous’ Angels.

Because in agentic crime intelligence, cyber forensics, OSINT, AML, and AI-assisted investigations, the dangerous failure mode is not “the model sounded dumb.”

It is the model sounding convincing while being wrong, unsupported, manipulated, overconfident, or legally radioactive.

SABLE is being designed as the adversarial governance layer between agent output and human trust.

Not another chatbot.

A circuit breaker.

The job is simple:

Before any finding becomes a case note, report, lead, accusation, or exported brief, SABLE attacks it.

Claim extraction.

Source-grounding.

Contradiction detection.

Prompt-injection screening.

Policy checks.

Confidence calibration.

Human-in-the-loop escalation.

Replayable traces.

Audit logs.

Evidence provenance.

Basically: “cute theory, now prove it.”

For Nous’ Angels, every Angel has a role.

VERA ingests evidence.

NOX reconstructs timelines.

MIRA maps entities.

LUX writes the brief.

But SABLE asks the ugly questions:

Where did this claim come from?

Which evidence supports it?

Which evidence contradicts it?

Is the source reliable?

Is the model hallucinating?

Did a poisoned document manipulate the agent?

Is this an investigative lead or an unsupported accusation?

Does this need human approval before it touches reality?

That is the difference between “AI said so” and defensible case intelligence.

The technical stack I’m researching for SABLE is very much governance-first:

@guardrails_ai for input/output validation and structured guardrails.

@LakeraAI for prompt injection and hostile-input defense.

@giskard_ai for red-teaming, hallucination testing, and adversarial evals.

@ragas_io for RAG/agent evaluation metrics like groundedness, faithfulness, context relevance, and answer quality.

@ArizePhoenix, @langfuse, and @traceloopdev for observability, tracing, eval logs, tool-call visibility, and replayable agent runs.

@OWASPGenAISec for the actual threat model, because agentic AI without a threat model is just a security incident wearing eyeliner.

The architecture I want:

Every agent action becomes a trace.

Every trace links to evidence.

Every evidence item has provenance.

Every claim gets scored.

Every risky output gets escalated.

Every report has human signoff.

Every mistake can be replayed.

That matters because crime investigation is not normal software.

A hallucination in a recipe app gives you cursed soup.

A hallucination in forensics can ruin someone’s life.

So SABLE is not there to make Nous’ Angels slower.

SABLE is there to make it survivable.

The future of AI investigation will not be won by the system that generates the most confident answer.

It will be won by the system that can show its work, expose uncertainty, survive scrutiny, and know when to drag a human into the room.

Not AI cops.

Not vigilante nonsense.

Agentic forensic infrastructure with a professional hater in the loop.

Five agents.

One case.

No loose ends.

[Daft.Exe @DaftExeOS]

@Memesrugged @VillianOfCode What other project is around after 7 months? We don't pander to you gamblers, stick around or don't.

[Daft.Exe @DaftExeOS]

I created the Nous' Angels infra and will be implementing it elsewhere as well - @VillianOfCode is one of a kind. IYKYK.

[Villano @VillianOfCode]

Nous' Angels — what's shipping next 🖤

The investigative substrate is live. Now we plug in the rest.

- Agentic depth (@NousResearch Hermes) • Persistent skill creation — Angels author & reuse investigative playbooks • Sandboxed code execution for evidence transforms (hash diffs, EXIF, blockchain calls) • Browser-automation evidence capture with provenance metadata • Multi-step tool chains written to the agent_runs audit log

- Retrieval & graph • Graph-RAG over case evidence (entity + timeline edges) • pgvector semantic search across OCR'd files, transcripts, threads • Force-directed entity graph with cited edges • Alias clustering UI with manual merge/unmerge

Stay updated = https://t.co/1HYAH9tOX9

@teknium appreciate if you could take a look at what we're building here!

[Daft.Exe @DaftExeOS]

@Kcain1982 @VillianOfCode Still here

[Daft.Exe @DaftExeOS]

@Kcain1982 @VillianOfCode Don't be daft ;)

[Villano @VillianOfCode]

Nous' Angels — what's shipping next 🖤

The investigative substrate is live. Now we plug in the rest.

- Agentic depth (@NousResearch Hermes) • Persistent skill creation — Angels author & reuse investigative playbooks • Sandboxed code execution for evidence transforms (hash diffs, EXIF, blockchain calls) • Browser-automation evidence capture with provenance metadata • Multi-step tool chains written to the agent_runs audit log

- Retrieval & graph • Graph-RAG over case evidence (entity + timeline edges) • pgvector semantic search across OCR'd files, transcripts, threads • Force-directed entity graph with cited edges • Alias clustering UI with manual merge/unmerge

Stay updated = https://t.co/1HYAH9tOX9

@teknium appreciate if you could take a look at what we're building here!

[Daft.Exe @DaftExeOS]

Gm.Exe

Working in silence as usual, we've been busy throughout April behind the scenes - trying to acquire a substantial partnership which will shape the future of $DAFT

Taking steps to overcome limitations - even though our minds have no limitations. The next few months...

[Anoncoin @anoncoinit]

This is going to be exciting 👀

[Daft.Exe @DaftExeOS]

HugeUpdateOTW.Exe 😏

Who's ready to Touch some Garden... 🍃

[Daft.Exe @DaftExeOS]

WeLike.Exe

[Low Latency Myth @LowLatencyMyth]

Debunking the Low Latency Myth: https://t.co/bynKtNX2Fb

We're setting a benchmark where developers looking into AGI will bottleneck their latency set-up with us.

Few understand right now.

[Low Latency Myth @LowLatencyMyth]

Debunking the Low Latency Myth: https://t.co/bynKtNX2Fb

We're setting a benchmark where developers looking into AGI will bottleneck their latency set-up with us.

Few understand right now.

[Daft.Exe @DaftExeOS]

While ARC-AGI-3 leaves the scaling bros with <1 % action-efficiency and their game-studio benchmark in existential shambles, $DAFT is already cultivating the actual substrate for AGI: the AGI Dream Garden.

Forget token-prediction theater. Here, human-trained custom models don’t hallucinate environments... they embody them. Each agent is instantiated as an avatar undergoing real-time perceptual transduction in a persistent, multi-agent simulated ontology. Sensory streams are grounded via continuous RLHF loops fused with avatar-specific reward models, creating qualia-grade experiential embeddings that scale with interaction depth rather than parameter count. These avatars don’t just monologue in latent space; they engage in decentralized inter-agent discourse protocols, emergent pidgins of vectorized theory-of-mind, forged through symbiotic model federation.
Custom fine-tunes bootstrap collective intelligence via federated gradient flows across the garden lattice: one agent’s novel policy becomes another’s inductive prior, turning isolated LLMs into a living, self-organizing superorganism. Zero-shot? Cute. This is experiential few-shot across lifetimes of avataric embodiment - where ontological grounding, causal abstraction, and recursive self-improvement arise organically from persistent world simulation + human-augmented preference alignment, not from yet another frozen weights snapshot.

The benchmark studio built a gauntlet.

$DAFT planted the garden where AGI actually grows.

It’s not “might be over.”

It’s already germinating.

[can @marmaduke091]

🚨 ARC-AGI-3 is released and the situation is crazy

&gt; No model gets more than 1%

&gt; "We achieved AGI" bros in shambles

&gt; They created an entire game studio to make this benchmark

&gt; Either way it's going to be solved before 2027

It might be over

[Daft.Exe @DaftExeOS]

Nurturing a new ARC. 🖥️

[Daft.Exe @DaftExeOS]

@DogeOS What a time to be Alive.Exe

[Daft.Exe @DaftExeOS]

Every decision we make is for a reason. Impatient humans - there isn't a place for them. 🫡

[AiBattle @AiBattle_]

ARC-AGI-3 launches tomorrow

- The first interactive reasoning benchmark built to test human-like intelligence in AI

- 1,000+ levels across 150+ environments requiring exploration, learning, planning, and adaptation

- Video-game-like tasks with no instructions, requiring multi-step reasoning and rule discovery

The highest score on ARC-AGI-1 currently is Gemini 3.1 Pro with 98%, while on ARC-AGI-2 it is Gemini 3 Deep Think with 84.6%