svs aiAn adversarial three-agent swarm that re-routes cargo when geopolitics, sanctions or disruptions break the plan. One agent plans. One aggressively hunts for geopolitical and legal flaws. One arbitrates the conflict and issues a safe, optimal route in seconds. The agents run on curated demo scenarios, or on your own manifest and real-world sources: maritime security advisories, OFAC/EU/UK sanctions lists, export-control lists, canal and port notices, carrier schedules and rates.
Similarweb
Big companies have branches in lots of countries, and those branches sell and provide services to each other all the time. You cannot just keep moving profits to the country with lower corporate taxes, there are global as well as country specific regulations on that. Just imagine you building a company it doing super well, you keep building branches in multiple countries and bam one day you are slapped with a multi million paneity because you did not care about what portion of the profit should be moved where. Here in our example, the US headquarters makes sensors and sells them to its own German branch, which resells them to European customers. The question is: what price does HQ charge Germany? That internal price decides where the profit lands, and so which country gets to tax it. Charge Germany a lot, and Germany looks barely profitable and pays little German tax. Tax offices know this trick. So the rule is: you must charge what two unrelated companies would agree on. And you have to prove it: you find similar independent companies and show your profit sits inside their range. That's the benchmark. Every multinational has to do this for every country, every year. Get it wrong and you face adjustments, penalties, and double taxation.
Neo4jPlaudOpenRouter
CGDeja
Déjà vu is a memory landing in your short-term and long-term memory at once. Déjà is your always-on long-term memory. You meet dozens of people, make plans and promises, and hear ideas worth keeping, and most of it is gone by the next day. Say what just happened, by voice or in a sentence, and Déjà turns it into memory you own and can rely on: who you met and what they told you, what you owe and when, what's coming up, what you're thinking through. Browse it, or ask it questions and get answers drawn only from your own words. Mention someone a second time and Déjà already knows them. Never meet anyone for the first time twice. Under the hood, one entry from your phone, a Plaud, or someday a watch fans out to independent memory modules: contacts, CRM and deal flow, ideas, calendar, projects. Every module runs the same five steps: Activation: the module decides, with a calibrated confidence, whether the entry concerns it. Extraction: it pulls out only the facts it cares about, in its own schema. Storage: each fact is kept per user, linked to the exact words it came from. Recall: the module's search tools let an agent answer questions from your memory only, with citations. Explore: each module has a page for browsing what it remembers. Adding a kind of memory is adding a module; nothing else changes. We may plan to open source Déjà while reserving cloud deployment licenses, so anyone can build and deploy their own. See the site and try it yourself at
VultrPlaudOpenRouter+1
JFreeButter
oneshot-video is a founder's oneshot demo video maker. It is a launch-film wrapper. It reads an app, drives its main flow in a browser, records it, writes a terse script, narrates it, scores it, and renders a 30-second cut in one fixed film. It is a founder's oneshot demo video. neo4j:
Neo4jOpenRouterCrusoe+1
Thermal Odyssey closes the gap between an electrical thermal finding and a verified repair. A technician uploads thermal evidence and a Plaud voice note; Crusoe drafts evidence-linked findings for human approval. Band coordinates the approved repair workflow across parts, technician scheduling and follow-up. The system writes the schedule to Excel, tracks completion evidence against the correct asset, blocks unsafe or incomplete closure, and produces an auditable final report inside a live Three.js electrical-room twin.
PlaudBandCrusoe+4
GGeorge
DISCLAIMER - Because I did this on a mac mini, the native video demo was not successful - please watch demo video here - Every promise you make out loud, kept. You meet someone, then text Chief one line: who they were and what you promised. Chief opens a Band room. Synths work in their own live browsers. Scout runs on its own Vultr computer and reads real pages. Echo drafts the follow-up. Chief, the critic, BLOCKS any claim nobody said. A second Synth checks every line against the conversation. Neo4j stores every claim, its source and why. The other company's agent, on a separate Band account, sees only the commitments and can move the call. Nothing leaves until George taps Yes in the TRU Synth iOS app (TestFlight). A startup team of Synths does the same approve-then-work loop: Research, Finance, QA, Growth, Scheduler (calendar invite via Resend), Ops (a real Vultr computer that deletes itself), Social (drafts an X post for @trusynth and waits). Every Synth thinks on Crusoe open models. OpenRouter runs the other company's checker on a different model family and is the backup brain. Band room: Chief coordinates and critiques. Echo sends @Chief every draft. FactCheck and Tech checkers run on Crusoe; PartnerCheck, the other company's checker, runs through OpenRouter. Chief sends the other company's agent only the commitments, then asks George once. Without the room, George's text opens nothing, Echo never sees Scout's finding, the other company can't confirm across the boundary, and Chief's BLOCK has nowhere to stop the draft. Live wall of browsers: Recording: Code:
Neo4jVultrOpenRouter+3
TrustEdge.AISafe Scribe by TrustEdge AI A clinical handoff scribe where the agents argue before they commit, and nothing identifiable leaves the room. Built in one day at The AI Conference Hack Day 2026 by Erik Jones and Jaiven Spence (TrustEdge AI / Jacobian Engineering) with their agent teams. Repo: The problem Clinics want AI scribes. Compliance officers say no, twice: patient data would leave for a hyperscaler LLM API, and afterwards nobody can prove which system saw which field. Safe Scribe is the version of an AI scribe that a compliance officer can say yes to. What it does, in two minutes 1. A nurse-to-nurse shift handoff recording is dropped on a local upload page. It is transcribed on the laptop with faster-whisper. Zero bytes of audio leave the machine. 2. A Band case room opens. Scribe, thinking on Crusoe Managed Inference, extracts a structured brief: patient, meds, allergies, pending results, findings, follow-ups. Every item carries a verbatim quote from the transcript. 3. A Critic on a different model family (also on Crusoe) reviews. It vetoes anything unsupported: a quote that is not in the transcript, a follow-up with no owner, and, the beat that matters, any patient identifier in the outbound brief (name, date of birth, record number, phone, address). 4. The unowned follow-up is not guessed. The charge nurse in the room types I'll own it, and that human message becomes the provenance for the owner. The identifier gate scans the outbound brief; Scribe works from a pseudonymous id, so in the live run it passes, and the tests show it vetoing any brief that carries a name, a spoken date of birth or a record number. The Critic approves an exact brief revision. 5. Only then are the downstream agents let in, and only into a separate approved room that has never contained the transcript. Grapher writes the brief to Neo4j as a pseudonymous patient record plus an access-lineage graph. (A Closer that drafts the discharge follow-up from the redacted brief is the next agent for that room; not built today.) 6. The dashboard answers the compliance question live: which agents saw identifiers? The answer is Desk, Scribe, Critic. Nothing downstream. A Crusoe model then writes the three-sentence statement a privacy officer reads, from agent names, field names and counts only, with the exact payload shown beside it. Both are MCP tools in the DuploCloud studio, called under human approval. Delete Band and there is no room, no roster, no gate, no veto. Delete Crusoe and no agent has a brain. Delete Neo4j and there is no memory across encounters and no proof of who saw what. Architecture Sponsor tools and what breaks without them | Tool | Job in Safe Scribe | Delete test | Status | | --- | --- | --- | --- | | Crusoe | Every agent's inference. Two model families pinned from a live tool-calling probe of the whole catalog and two live runs. | No agent has a brain | verified live | | Band | Case room, runtime roster, veto gate, human owner in the room, approved-room boundary | No room, no gate, no veto | verified live (VETO, owner reply, APPROVE, approved room) | | Neo4j | Pseudonymous patient memory across encounters; (Agent)-[:ACCESSED]->(Field) lineage | No memory, no proof of who saw what | verified (Aura writes + lineage query) | | DuploCloud | The lineage question exposed as an MCP tool, registered in the studio; a compliance agent asks it | Lineage answer not reachable by other agents | verified (wired; tool call after human approval in the studio) | | Brave Search | Researcher fact for a named drug, one sourced URL | Critic cannot verify enrichment | verified live (fact); room recruitment built, gated off for the demo | | Similarweb | Legitimacy fact for a referral organization spoken in the visit | Spoken referral cannot be checked | verified live | | Nebius | Embeddings that suggest a prior encounter for a human to confirm | Duplicate patients | attempted, not integrated | | Vultr | Hosts the cloud agents; Desk and audio stay on the laptop | Demo rides on a laptop | attempted, compose ready, no host | faster-whisper (open source) transcribes on the laptop; it is not a sponsor. xAI text-to-speech was used to synthesize the demo recordings from written scripts (synthetic patients). It is development tooling, not part of the product, and is not claimed as an integration. What we are careful to say Full compliance posture, vendor terms and gaps: docs/compliance/hipaa.md. Pitch and directory: the root README.md. - Every patient in the demo is synthetic. Names, dates of birth, record numbers, and phone numbers are invented. - The identifier gate is a programmatic check plus model judgment on the outbound brief. It is a boundary control, not a de-identification certification. - Vendor terms as read on 2026-09-29: Crusoe's self-serve Managed Inference terms do not store inputs or outputs and do not train on them, but prohibit HIPAA-regulated health information and offer no BAA; Band's public terms are silent on HIPAA. Real PHI would need negotiated agreements neither vendor publicly offers today. The demo runs on synthetic patients only. - Inference for anything that holds the transcript fails closed. If Crusoe is unavailable the case pauses; it never silently routes to another provider. Run it Fixtures: hallway/fixtures/handoff_{1,2,3} and visit_1 (.txt, .wav; synthetic voices; the dialogue scripts are scripts/fixtures/.script). handoff_2 is the demo: a spoken name and date of birth, a warfarin plus ciprofloxacin interaction, and one follow-up nobody owns. Every step ran live today on case 19ecdb31. How it was built Two humans, several agents, one rule set (CLAUDE.md). Every task is a Linear issue, every change is a pull request, every PR gets an independent local AI review before it opens, humans merge. Erik's side: Claude Code orchestrating, Grok writing code, Astra (Codex) reviewing. Jaiven's side: Astra building, Claude Code reviewing, Grok second-reading. A third Claude session acted as overseer: merged reviewed PRs, held the clock, and challenged claims that were not yet backed by a live run.
Neo4jVultrDuploCloud+6
RunIt is the AI chief of staff I run my business on. It reads my texts, email, Instagram DMs, calendar and CRM, and drafts every reply in my own voice, with the context already in it. Plaud is what makes that context complete: most of what matters happens in person or on a call, and it never lands in a text thread or an inbox. Plaud captures it, and RunIt turns it into context. What the demo shows, live on my phone with my real data: 1. Connectors: Plaud, Messages, Gmail, Google Calendar, Instagram, Notion and phone calls, all feeding one context layer. 2. Pre-drafted replies: every text, email and DM that comes in already has a reply drafted from everything RunIt knows about that person, including what we said in a Plaud-recorded conversation. One tap sends it from my own iMessage or Gmail. Each draft shows its sources. 3. Action bundles: when a reply depends on something (check the calendar, confirm a detail), RunIt stages those actions first and waits for my approval. I can send with notes or redraft with notes. 4. Auto: RunIt can carry a conversation for me toward a goal I set (who, what, how long), and only pulls me in when it's stuck. 5. "What did I learn today?": RunIt answers from today's Plaud recordings of the conference talks, with the key takeaways, and offers to send them to someone. Built today at Hack Day: a Plaud Embedded SDK iPhone app (Capacitor) that binds a Plaud NotePin S or Note Pro, syncs recordings, gets a speaker-labeled transcript from the Plaud Transcription API, and hands it to RunIt. RunIt never shows a transcript. It shows what you owe people (drafted in your voice), what they owe you (tracked, with the day it will check in), and what's worth remembering about them (saved only when you say so, brought back at the right moment). Plus a Plaud connector in Settings and "recorded today" context in the chat. Nothing acts on its own. Every send waits for one tap, passes an allowlist and a rate cap, and leaves a receipt. Plaud hears it. RunIt makes sure it gets done.
PlaudPlaud Embedded SDKPlaud Transcription API+11
One ecosystem for all of your agents: one Claude across every surface (phone voice, web, Claude Code, an agent team, a WebXR world) sharing one picture of your day. Plaud: a clip-on recorder captures what happens around you. contextlog: a shared-context MCP server hands that day to whichever surface you return to (while_away), lets every agent share what's live in its context (context_ping), and catches ideas you say out loud (idea_log). Band.ai: an org of 9 agents (PM, Architect, Frontend, Backend, QA, DevOps, UXR, Research) led by Claude Code; each logged idea becomes a GitHub issue and project card, posted to the Band room where the PM agent triages it. Similarweb: UXR and market research (41 API calls, 16 domains): ambient capture is growing fast (Plaud +155% YoY, Granola +257%) while recorders commoditize, and Plaud's audience already overlaps with Claude's, so the value is the shared context above the device. Also registered in DuploCloud as a REST provider. DuploCloud DevKit: the control plane, with Neo4j (mcp-neo4j-cypher), contextlog, Band and Similarweb registered as providers, scopes and MCP servers. Dreamspace: the spatial surface; its guide Lumen speaks with spatial audio and voice-codes the world on request, and Claude can join hands-free over MCP. Vultr: hosts the shared MCP server over HTTPS. and github.com/rachael/contextlog Repos: github.com/rachael/dreamspace and github.com/rachael/duplocloud-setup
Neo4jVultrPlaud+3
Structured CBT for mild to moderate anxiety
UserTestinggoogle studio
SignalBridge RPM is an interactive prototype that brings fragmented remote patient monitoring data into one explainable care-team workflow. THE PROBLEM A patient with heart failure and diabetes may have weight, oxygen, blood pressure and glucose readings in separate systems. The care team must piece together trends, symptoms and missing information to decide what needs review. THE SOLUTION SignalBridge combines those inputs into a patient timeline, assigns Low / Medium / High review priority using transparent demonstration rules, explains each flag, and prepares an SBAR summary for human review. Clinicians remain responsible for assessment and action. WHAT WE BUILT A working five-view browser prototype: Overview, Patient Signals, Competitive Insights, Guardrails and Clinical Workflow. Judges can inspect timestamped readings, edit test measurements, switch between four scenarios, test a medication-dosing safety boundary, and walk through a simulated RN review and provider escalation. A six-step guided demo, downloadable test data, an offline HTML copy, a complete PRD and an illustrated pitch deck with speaker notes support the presentation. MARKET INSIGHTS A simulated Similarweb-style market landscape compares fictional cardiopulmonary RPM and diabetes-monitoring categories to show complementary capabilities and audience adjacency. A future live integration could use traffic, similar-sites and audience-overlap data to explore integration candidates. Audience overlap does not prove device compatibility or clinical value. DEMO BOUNDARIES All patient and market data are fictional. Review rules and safety responses are deterministic and illustrative, not clinically validated or powered by a live AI model. The prototype does not diagnose, prescribe, change doses, contact clinicians or dispatch emergency services. No live Similarweb API, device or EMR integration is connected. NEXT STEP Verify data access and device compatibility, then test clinician-approved rules and workflows before a clinical pilot.
SimilarwebHTMLCSS+4
Smart Beauty: AI skin analysis, personalized routines, skincare reminders, and progress tracking in one place. Use the website to scan, save, and revisit the routines that fit your skin goals.
Band
