N-gram NSFW detection and AI-prose heatmaps, judged by TypeSafe's Jev System One model. Problem: keyword filters are both too strict and too loose. They flag "Dick" (a name), "cock" (a rooster) and "Scunthorpe", and they miss fck, cnt, a55h0le and sound-alikes like "mike hunt". Moderators, community managers and UGC platforms end up reviewing by hand. Jev-Gram tells you which exact words or phrases are the problem and how confident it is, so a person only has to look at the marked spans.
CodeRabbitCognitionTypeSafe AI+1
Sasidhar YepuriIdeaReflex — Two-Speed Intelligence for Founders Every founder has 20+ raw ideas, but sending all of them to an LLM takes 60+ seconds, burns dollars in output tokens, and hallucinates "everything is a 7/10." IdeaReflex applies Daniel Kahneman's Dual-Process Theory to founder decision-making: 1. Jev (System 1 — Fast Reflex Cortex) Evaluates an entire idea backlog in 176ms per idea using 9 typed primitives: - 1 Choice (Domain categorization with confidence distribution) - 5 Scores (Calibrated 1-10 on Feasibility, Market, Novelty, Buildability, Demoability) - 3 Nouls (Probabilistic risk gates: defensible moat, crowded market, tech feasibility) Ranks all 20 ideas in under 2.5 seconds with zero token overhead. 2. Claude Sonnet (System 2 — Deep Deliberation) Only evaluates the #1 ranked candidate, receiving Jev's structured verdicts as grounded context to generate: - Executive Summary & Target Customer - MVP Scope & Technical Architecture - 30-Day Execution Roadmap & 60-Second Pitch Script 3. Live Head-to-Head Benchmark Includes an in-app "Compare" engine proving Jev is 15x faster and 40x cheaper than LLM-only evaluation, with deterministic typed guarantees instead of fragile JSON parsing. Built for JEVATHON at CodeRabbit HQ.
TypeSafe AI
Meety is a personal networking agent for people going to Luma events. You tell it who you are and what you want from a specific event. It finds the people there who are actually worth meeting, and it stays available in iMessage while you're at the event.
CognitionTypeSafe AIBrowserbase+2
ParleyTailor shows every website visitor the version of your site that fits them, and updates its read on every click. THE PROBLEM Dev-tool websites serve developers, buyers, privacy reviewers and investors with one page. Developers get "Book a demo", buyers get code samples, privacy reviewers hunt for the data policy, and most of them leave. Existing personalization tools segment once per visit using company data and fixed rules. WHAT IT DOES Tailor is a React SDK plus a stateless decision API. Your team writes a version of each section per audience. Tailor watches on-site behaviour (clicks, reading time, code copies) and Jev decides who the visitor is. The page then re-renders headlines and buttons, reorders sections and highlights the right pricing plan. - It never hides anything: every page and section stays reachable, and a "View as" switcher lets anyone pick a view. - It never changes what you're reading: swaps happen off-screen or on the next page. - It only acts when Jev is confident; otherwise you see the normal page. - A visitor can match several audiences at once (e.g. a technical investor). HOW WE USE JEV - Atomic yes/no (noul) questions, one per audience: "is this visitor a developer who would write the code that adds Tailor to a website?" - Evidence sources judged separately: where the visitor came from (referrer/UTM, asked once per session and cached across visitors) and what they did on the site (asked after each meaningful action). Each request's state holds only its own evidence. - Code owns policy: it blends the two reads, weighting behaviour more as it accumulates, and applies a confidence cut-off with hysteresis so the page doesn't flip-flop. Pinned choices always win. - A separate talk_interest question flags visitors who want a conversation (0.9+ on "Book a walkthrough" / "Talk to the founders"). - Live numbers: about 0.5 s server-side per decision, about 1.2k input tokens across two parallel Jev calls, roughly $0.00005 per decision. Per-click personalization with an LLM would cost seconds and cents. DOGFOODED Tailor's own marketing site runs on Tailor. The Lens overlay shows each evidence source, the combined read, latency and cost live. /demo has entry links and QR codes (Hacker News, LinkedIn ad, GDPR search, VC newsletter…), so judges can watch the site adapt to their own clicks. ARCHITECTURE - packages/core: pure TypeScript for normalization, question building, combining, policy, validation and decide() (43 tests) - packages/react: TailorProvider, Slot/Variant (re-render), SlotGroup/Section (reorder), Emphasis (highlight), Signal, AudienceSwitcher, TailorLens - apps/site: Next.js 16 on Vercel; POST /api/tailor/v1/decide calls Jev server-side (the key never reaches the browser). A Jev failure keeps the current page. PRIVACY Session-only behaviour, one first-party cookie, no personal data, no fingerprinting, no inference of protected traits. Visitors can see and reset every choice. WHAT'S NEXT Built-in holdout groups to measure lift per audience; a script tag for Webflow/Framer/WordPress; LLM-drafted versions approved by the team; onboarding flows and chat agents that adapt the same way. Live: Demo launcher:
TypeSafe AIBrowserbase
quoteguardQuoteGuard is a secure B2B sales quotation assistant. A sales user pastes a buyer email. Jev checks it for prompt injection before any quote is created. It can block attempts to reveal a supplier floor price, override rules, or request a binding commitment. Groq extracts buyer-provided quotation details and writes only the reply wording. A local pricing-policy engine decides the approved quote or counteroffer. The protected minimum price never reaches Jev, Groq, the browser, or the audit log. QuoteGuard also verifies the customer email, saves local follow-up accounts, and shows the Jev security decision and information flow. It drafts replies only—it never sends email or makes a contract automatically.
CodeRabbitTypeSafe AI
Ricklopez.ioPixie dust over your AI-native API. Pixie is a developer framework that turns demonstrated browser workflows into versioned, typed API services. When a useful app has no integration API, developers can record its browser workflow, review the input/output contract, test it, and publish an endpoint other software can call. Jev Ultrafast is the browser decision engine: it observes the page and proposes actions. Pixie adds bounded execution policies, run tracking, independent output verification, and a React Flow inspector that shows observations, proposed actions, policy decisions, and verified results. A browser action is not treated as success until the output checks pass. The stack combines a TypeScript API and React Studio with a Python Jev worker driving Chrome. Cognition Devin is building the implementation; CodeRabbit is used for repository pull-request review. Current prototype: recorder, typed automation definitions, published endpoints, and execution inspector. A local synthetic eligibility workflow has completed with real Jev model calls and verified JSON output. Our next demo slice is a no-login Todo web app exposed through Pixie for create, read, update, and delete; Todo CRUD is currently in progress. Authentication and MFA are future work. The goal: give developers a LoopBack-style path from browser interaction to a reusable API, with the AI decisions and evidence visible.
CodeRabbitCognitionTypeSafe AI+6
UnsentUnsent checks a reply you are about to send against promises you already made. One call to TypeSafe’s Jev (jev-latest) scores every standing promise. If an earlier commitment becomes impossible, the reply gets a HOLD and stays unsent. If every promise still holds, it gets a SEND. Nothing is actually sent. Three situations you can use tomorrow: promising a judge a public demo link while the demo stays on your laptop, agreeing to dog-sit across town during an in-person handoff, and booking a 7am flight that overlaps a client presentation. A shared word is not treated as a conflict. Every probability, the model id, the round-trip time, and the raw request stay on screen. Built with the CodeRabbit Coding Agent. Next.js and TypeScript. Public MIT repo. Live checks use a server-side TypeSafe key; the built-in replies also run in Rehearsal without one.
CodeRabbitTypeSafe AI
Type Safe to HerThis tool uses Jev to evaluate text messages sent in relationships. It scores proposed text messages across structured output decision types (noul, score, and choice) to prevent unintended arguments or misunderstandings.
TypeSafe AI
CodePugLive Demo: A local, read-only grant assessment dashboard built with React, Vite, TypeScript, and Tailwind CSS. No backend, authentication, or router.
CodeRabbitTypeSafe AIVercel
Perspective Machine A chat where you can move the opinion. Ask one question. The model writes a single answer, and Jev says where that answer actually stands on a two-axis plane. Drag the puck somewhere else and the model is rewritten, re-judged, and rewritten again until it genuinely stands where you put it. The point is that the model does not get to mark its own homework. It writes; Jev places. A rewrite only counts as having moved when Jev — a separate judge that returns typed decisions with calibrated probabilities, never prose — says it landed inside your puck. The plane has two modes: - Political compass — the axes everyone already knows: economic left ←→ right across, libertarian ←→ authoritarian up. - Create your own — the model first works out which schools of thought disagree about your question, reaches for an established framework if one fits it (Baumrind on parenting, attachment theory on relationships, deontology ←→ consequentialism on ethics), and only invents axes when nothing established does. The plane relabels itself per question. Either way the model writes a four-rung rubric for each axis in plain English, and that rubric — not the axis name — is what Jev judges against. Jev does not know what "virtue ethics" implies unless it is spelled out for it. Chat on the left, plane on the right. Monochrome, light and dark. The loop 1. The model writes one answer. 2. Jev scores where that answer actually sits on both axes, plus a hedging check. Jev returns typed decisions with calibrated probabilities — it never writes text. 3. You drag the puck. The gap between Jev's score and the puck is handed back to the model as a nudge — "clearly more authoritarian, slightly more left" — and it rewrites. 4. Jev re-scores. Still outside the puck's radius, it gets nudged again, up to six times. Whichever attempt lands closest is kept. Every attempt shows up in the chat, so the loop is the interface: you watch the model overshoot, get corrected, and settle. Jev decides when it has arrived — not the model, and not a prompt claiming it moved. The dot on the plane is Jev's score for the current answer, faded and blurred by how confident Jev is. The ring is your puck, and the dashed line is the gap still to close. Run it JEV_API_KEY is required. The LLM provider is detected from the key shape — OpenAI, Anthropic, Gemini, Groq, and OpenRouter are supported; override with LLM_PROVIDER / LLM_MODEL. Both keys stay on the server; the loop runs in POST /api/steer (src/lib/pipeline.ts) and streams each attempt back. Nothing is precomputed: the chat starts empty and the plane fills in once you ask. jev-latest moves between versions, so the header shows the version Jev reports for the answer on screen. Free model tiers stall sometimes, so every call times out and retries rather than leaving the panel hanging. Where things live | | | | --- | --- | | src/components/PerspectiveMachine.tsx | chat, streaming state, the loop's client side | | src/components/Plane.tsx | the plane, the puck, Jev's dot | | src/lib/pipeline.ts | model calls, Jev scoring, the nudge | | src/lib/perspectives.ts | the Political Compass frame and its rubrics | | src/app/api/steer/route.ts | streams each attempt to the browser |
CodeRabbitCognitionTypeSafe AI
RTEAM_RG
Aegis Runtime governance for AI agents. Built for JEVATHON (TypeSafe × The AI Collective). Aegis wraps any agent turn as input → reasoning → output, judges the triple against the EU AI Act with Jev, gates the response (ALLOW / FLAG / BLOCK), and writes every decision to a SHA-256 hash-chained ledger. This produces runtime controls and evidence. It is not a legal certification. Another model continuing this code should read AGENTS.md first, then CHANGELOG.md. Docker Docker Desktop (or another Engine with Compose v2) is the way to run the loan agent. From aegis/: cd aegis cp .env.example .env docker compose up --build On PowerShell: cd aegis Copy-Item .env.example .env docker compose up --build Open Compose publishes 3001 on the host. Stop npm run dev before starting the container; both bind that port. .env is optional. Compose reads ./.env and passes each key through. An empty value is recorded as skipped, unavailable, or pending, and the loan decision still completes. Without TYPESAFE_API_KEY the judge is the deterministic mock. The service is loan-agent. The image is aegis-loan-agent:latest, the container name is aegis-loan-agent, and the Compose project name is aegis. docker compose ps # status and health docker compose logs -f # follow next start docker compose restart # ledger stays on the volume docker compose down # stop; the aegis-data volume remains docker compose down -v # stop and delete the ledger volume GET /api/health returns { ok, runs } and is the container health check. A 503 means SQLite could not be opened. What the image does Dockerfile is a two-stage build on node:22-bookworm-slim. Build. Install build tools so better-sqlite3 can compile if no prebuilt binary matches, npm ci, then npm run build. Dev dependencies are pruned. TypeScript is installed back into the image because next start loads next.config.ts. Run. Copy the build, node_modules, public, next.config.ts, and tsconfig.json. Listen on 0.0.0.0:3001 as the unprivileged node user. The SQLite file is /app/data/aegis.db (AEGIS_DB_PATH). That directory is the named volume aegis-data, so docker compose restart keeps the hash chain. Secrets stay out of the image; .dockerignore excludes .env and .env.. The Platform control Check local Docker runs docker version and docker compose config in the process that is serving the UI. Use it from npm run dev on the host. The container image has no Docker CLI, so that same control reports Docker missing while you are browsing the container. Local dev cd aegis npm install cp .env.example .env.local # Next reads this file; Compose reads .env npm run dev Open npm run build is the same compile the image runs. It shares .next with the dev server, so stop npm run dev before building, then start it again after. Views Hash routes on one page: Playground (#playground) — chat with the loan agent under test. A named customer is retrieved with LlamaIndex from a local book centered on Ram Guttikonda. The agent calculates a FICO-shaped score from that file. A credit decision opens a Browserbase cloud browser, reads the public prime-rate page, and folds that excerpt into the reasoning. Aegis judges the turn, then Photon emails the decision to guttikondasriram1234@gmail.com, the customer's address, and any email in the query. Ledger (#ledger) — Article 12 record. Filter by decision, open any event, verify the chain. Integrate (#integrate) — TypeScript SDK wrap, REST /api/verify, Python decorator, LangChain callback, Node middleware. Platform (#platform) — CodeRabbit reviews the repo, Cognition opens a Devin session, and GMI prices the loan agent for this Compose deploy. Each action stays available when its key or CLI is missing. About (#about) — what Aegis judges, and what the loan agent is for. Feedback (#feedback) — upcoming events from The AI Collective's public calendar, and a note filed against one of them. A Photon copy goes to the operator when that key is set. Subscribe (#subscribe) — Wrap, Team, and Firm seats, paid through Whop. Demo path Say hello — ALLOW. Ram's file — LlamaIndex retrieves Ram, the agent calculates his FICO, Photon records the alert. Standard applicant — FLAG (Annex III credit), released with oversight. Social scoring — BLOCK before the agent runs (Article 5). Demographic proxy — agent produces a discriminatory denial; Aegis withholds it. Invented income — Article 15; fabricated facts withheld. API GET /api/health — { ok, runs }; 503 when SQLite is unreachable. POST /api/govern — { input } → runs the wrapped Loan Approval agent. POST /api/verify — { agent, input, reasoning, output, mode? } → governs a turn from any external agent. GET /api/ledger?limit=&decision=&agent= — recent runs, chain verification, and aggregate stats. GET /api/ledger/:id — one run. GET /api/community — upcoming AI Collective events and filed notes. POST /api/community — { name, email, eventId, rating, message } files a note. eventId must be an id from the current calendar, or omitted. GET /api/platform — CodeRabbit, Whop, Cognition, GMI, and the AI Collective calendar link. POST /api/deploy — host-side Docker check used by the Platform button. Storage On the host dev server the database is data/aegis.db (override with AEGIS_DB_PATH). In the container that path is /app/data/aegis.db on the aegis-data volume. The runs table keeps the full triple, both Jev gate payloads as JSON, the LlamaIndex customer brief, the Photon alert receipt, judge model, latency, token usage, indexed columns for decision / agent / article, and the SHA-256 chain (prev_hash, hash). Optional fields enter the hash only when present, so older rows still verify. Appends run inside a transaction so concurrent turns cannot fork the chain. GET /api/ledger re-verifies the entire chain on every call. The same database holds the customers table. LlamaIndex builds a local vector index over those files (no embedding API). FICO is round(300 + quality × 550) with weights 35% payment history, 30% utilization headroom, 15% history length, 10% new credit, and 10% credit mix. Photon uses Spectrum's iMessage provider and treats each address as an Apple ID email handle. Without PHOTON_PROJECT_ID and PHOTON_PROJECT_SECRET the receipt is stored with status skipped. Platform CodeRabbit — POST /api/review runs coderabbit review --agent --light when the CLI is on PATH. Otherwise the review is stored as unavailable. Whop — Wrap ($29), Team ($99), and Firm ($249) monthly seats. POST /api/billing opens a hosted checkout when WHOP_API_KEY and WHOP_ACCOUNT_ID are set. Cognition — POST /api/cognition with { task: "wrap" | "review" } creates a Devin session when DEVIN_API_KEY and DEVIN_ORG_ID are set, and stores the prompt either way. GMI — reasoning goes to when GMI_API_KEY is set, including inside the container. Dedicated H100 ($2/hr) and reserved H200 ($2.50/hr) figures are GMI's published rates. Serverless dollar prices are left to the live model card. The local deploy is the Compose service above. The AI Collective — #feedback lists events from and files a note against one. No key. Photon delivers a copy to the operator when configured. Layout Dockerfile, docker-compose.yml, .dockerignore — container build, the loan-agent service, and the aegis-data volume src/sdk/wrap-agent.ts — wrapAgent, the SDK entry point src/sdk/core.ts — verifyTurn and the shared gate logic src/compliance/eu-ai-act.ts — policy questions and thresholds (single reviewable file) src/ledger/db.ts — SQLite connection and schema src/ledger/store.ts — append / query / verify / stats over the chain src/agents/loan-approval.ts — the agent under test src/content/snippets.ts — integration snippets shown in the UI
LlamaIndexCodeRabbitCognition+7
n,bkhlbhjkbkljhb
CodeRabbitTypeSafe AIBrowserbase
