Skip to main content

Product Finder

The map of everything we have going on. Organized by the Engine / Forge / Lab / Backlog framework — the tier sets the posture. Engine is paying customers and cash-generating relationships (highest bar). Forge is calculated bets with identified customer zeros (building toward Engine). Lab is speculative work, games, and research bets (creative latitude). Backlog is parked / pre-Lab ideas with no active investment — waiting on a champion, customer zero, or strategic moment to graduate.

Engine has a zero-line. Some Engine rows (TwinSpires, Exacta) are services revenue that doesn't compose the six Deep Agents ingredients directly — cash without recipe. Still Engine, still highest bar.

For the deeper take on each ranked bet — what are we asking you to believe, what does it look like if everything goes right, and where does it sit in the sequence — see the Three Lenses framework. That page also synthesizes the strategic re-frame this catalog implies: the bottleneck isn't ingredients, it's distribution.

How this page works. Quick-scan table first — every product, one row. Detail sections by tier below. When a product accumulates real content (full spec, GTM material, architecture doc, dedicated roadmap), it graduates to its own page and this index links to it. Earned, not assigned.

Status legend. ✅ live · 🚧 in flight · 🌱 in concept / not yet started · ⏸️ paused · 🛠️ rebrand or pivot in progress

Quick scan​

ProductTierWhat it isStatus
ExactaEngine (zero-line)Gaming software vendor — French port + Windows→Linux migration + Charitable Gaming HHR extension. Cash without direct six-pillar composition.✅
TwinSpires — BeeHiveEngine (zero-line)Player Services platform for Kentucky Derby 152. Services revenue; not composed of the six pillars.🚧
TwinSpires — OneBrisEngine (zero-line)E-commerce + Elavon integration + catalog manager (CDK infrastructure). Services revenue; not composed of the six pillars.🚧
TFG / PolarisEngineThe Futurum Group's analyst intelligence platform (Decision Maker, Signal, BEV, Companies, 17 practice areas)✅
IonQ / Quantum (Bucket 1)EngineForward-deployed quantum AI/agent — field-sales briefcase + data lake🚧
🐝 Bike4Mind (the platform-product)EngineThe AI orchestration platform sold at bike4mind.com. 72+ models, 9 providers, RAG, multi-tenant. Encapsulates B4M CLI, Deep Agents, Project Intelligence, Tavern, and the rest of the kit. Makes $ by definition. Promoted Forge→Engine 2026-05-18. Full anchor page →✅
Agentic Research (was IonQ Bucket 2)ForgeAgentic paper writing/reproduction + Jupyter (Q/Work, Iter-QAOA). Leverages Bucket 2 aspirations — we think IonQ will like agentic paper reproduction.🚧
OptiHashiLab #3The quantum optimization product. “ChatGPT for business optimization.” Patent in progress. A Lab play for Erik.🚧
DFALLab #7Discover First, Apply Later — quantum-computing research bet. New Lab item.🌱
B4M CLI + ReAct kernelForgeNao's PR #5896 — Claude Code-style CLI on B4M. The kernel everything else builds on.✅
B4M MobileForgeiOS + Android apps. First mobile AI cognitive workshop on market.🚧
B4M Desktop AgentForgeClaude Code as desktop bridge. B4M agents reach local files.🌱
B4M Teams / EnterpriseForgeThe packaged SKUs — team collab + enterprise AI orchestration🚧
Slack AgentBacklogNatural language → Jira / Confluence / GitHub. Enterprise entry point. (PR #4884)✅
LaminarMind (aka B4M Project Intelligence)BacklogAnother agent-orchestration layer. Customer zero: Exacta. (PR #5634)🌱
Notebook CurationBacklogTransform AI conversations into shareable, citable knowledge artifacts.🌱
MementosForgeHot / warm / cold tiered memory and context system for agents.🚧
ArtifactsForgeCode + document artifact creation, versioning, and sharing inside B4M.🚧
Data Lake WizardForge11-phase guided flow for creating a data lake. Shipped inside Polaris; generalizes to all data-lake products.✅
QuestMasterForgePersistent agentic quests, My Quests dashboard, cross-session. (PR #4885)🚧
ToolkitsForgeComposable tool bundles (Research, Code, Design). James owns. (PR #5016)🚧
B4M PiForgeProject management agent + SRE loop. Kevin as customer zero.🚧
B4M SRE BotForgeKen's full-loop agentic SRE flows on B4M itself. Already triaging issues.✅
WorldVisionBacklogMobile camera → AI visual understanding. Game modes (Chess, Tarot, D&D). Robotics API.🌱
Magic Mirrors (Phantasia)ForgePhase 0 physical AI — B4M-powered smart displays. Q3 2026.🌱
Mobile MirrorsForgePhase 1 physical AI — mirror + wheels. Q4 2026.🌱
Full RoboticsForgePhase 2 physical AI — add manipulation. 2027.🌱
BedrockNews.comForgeNews done reliable, authentic, readable, clean. Rebrand of VibesWire.com — just registered.🛠️
NovelClaims.comForgeScience / STEM sibling of BedrockNews. Agentic paper reproductions from arXiv.🌱
StocksAndVibes.comForgePractical stock + options trading with LLM augmentation, paper trading, education. Was VibesTrader.com. Michael Conard.🛠️
HordeForgeAccounting + invoicing + tax + cap-table system. Will power Starlight Pediatrics. “Tessera” pro-forma engine.🚧
OrionBacklogB4M's GTM hunter agent. Built on B4M, sells B4M. (See Orion spec)🌱
🐉 HydraForgeWhite-hat security training tool. Used on Lumina5, Polaris, VibesWire. Being presented to TwinSpires this week — see tactic.🚧
Polaris CYAForgeChoose Your Adventure for Polaris. Prototypes at babelfox.com / deepak-cpo.netlify.app. Next mega-epic.🌱
LibreOncologyForgeMedical education platform. Designed; build paused for V3 component refactor.⏸️
⭐ StarlightForgeOne of B4M's bigger Forge bets. Direct-Primary-Care EMR for pediatricians, competing against Atlas.MD. Customer zero: Dr. Yogini. AI-scribe + parent app + synthetic personas are the wedge. Security verified by Hydra (0 CRITICALs). 7-sprint timeline to displace Atlas.✅
K2KanjiForgeThermodynamic-inspired spaced-repetition for Japanese, Korean/Hangul, Rubik's Cube, musical notation, cocktails, more. DAU leader. Promoted Lab→Forge 2026-05-18.✅
PotionQuestLab #1RPG with AI players playing AI-made stuff. Has a full RPG API.🚧
TavernLab #2Game (FPS-style). Frame rate is incredible. PotionQuest API integration next.✅
ChessLabChess client / experience.🚧
GoPettoLab #4GoPets 2.0 — virtual pets + strong AI + mobile AR cozy RPG game. The original concept reimagined on the modern agentic stack.🌱
Mythos / Agentic SecurityLabWhite-hat security agents for B4M. Inspired by the Mythos/Glasswing lockout incident.🌱
AegisLab (unranked)Hydra inverted — agentic vuln sweep across the top-N OSS packages, responsible disclosure as the output. Parked pending ROI clarity.🌱
AgentCFOLabCFO agent — speculative project mentioned in roadmap.🌱
Agent MarketplaceLabLong-term vision — agent discovery and marketplace built on B4M. Vision doc exists in lumina5.🌱
🪐 MarsWiki / MillionOnMars V2 (Civ4x.com)Lab #5Treated as one project (2026-05-18). MarsWiki = bespoke knowledge lake (MoM + space/STEM); MoM V2 / Civ4x.com = civ-4X game built on top. The substrate and the surface, paired.🌱
HavenBacklogDigital immortality through PGGI (Pretty Good General Intelligence). Built on the private-memory work in ~/Desktop/Ember. Re-parked to Backlog 2026-05-18.🌱
Bacteria WarsLabReal-time strategy game — you play as a bacteria.🌱
Orc HunterLabFirst-person Ukrainian-war simulation.🌱
D&D Room (WorldVision use case)Lab #6Physical-AR D&D session: 8K monitors around the room, tablets and mobile phones, smart table, beanbag toss / axe throwing / darts as in-world mechanics, thousands of bespoke function calls and tools, agentic AI running the whole game. The flagship demonstration of what WorldVision + B4M Deep Agents can deliver.🌱

Engine — paying customers​

Exacta​

Gaming software vendor relationship led by Nimai Malle. The current 2026 work is twofold:

  • French localization port of the legacy Exacta system (formal deadline May 15, stretch April 20)
  • Windows → Linux migration to stop the Windows licensing bleed
  • Charitable Gaming HHR extension — new product line Nimai secured funding for; 5–6 month push. Kevin leads with stronger PM and months-ahead tasking. (From 2026-02-09.)

Where it shows up. Apr 4 update · Feb 9 update

TwinSpires​

Horse racing platform. Two distinct product lines under one customer relationship.

BeeHive​

One-pane-of-glass Player Services platform for Kentucky Derby 152. Marketing Ops surface. Includes the Create Bonus epic (Marketing Ops facing). Strong team velocity. Hard goal: Player Services 100% on BeeHive (not Boss) by Derby 152.

OneBris​

E-commerce + Elavon integration + catalog manager + frontend kept current against latest feedback. Migrated from SST to CDK infrastructure-as-code in close partnership with the TwinSpires platform team. Hard goal: some portion of OneBris alive and selling PDFs by Derby 152.

Team. Angelo, Victor, Rexen, Ken, Kyle, Wilfred, Chad, Gladys.

Where it shows up. Apr 4 · Feb 9

TFG / Polaris​

The Futurum Group's analyst intelligence platform — Polaris. The largest single product B4M operates. Scale: 902,685 LOC across 5,249 files, 157 database models, 743 API endpoints, 36 AI models across 4 providers, 129 callable tools, 17 active practice areas. See The 14 Pillars for the depth case.

Major features and surfaces:

  • Decision Maker — proprietary survey data on executive buying intent
  • Signal — vendor evaluation framework with LLM scoring
  • BEV Calculator — Max's ongoing feature work
  • Always Live Market Model — Jason's push, with 4 parallel experiments converging
  • Companies — canonical entity model with 100+ fields per company
  • Intelligence Feed — real-time competitive intelligence aggregation
  • CYA prototypes — the next mega-epic (see Forge below)

2026 posture. Stabilization surge (latency, P1 bugs, UX/feature debt) — the company-wide meta-P1 for the spring. Then BEV, Signal, and Always Live mature. Then CYA.

Where it shows up. Apr 4 — the surge · Feb 9 · 14 Pillars

IonQ / Quantum — Bucket 1​

A Polaris-Briefcase-style data lake + bespoke tools + agentic experience to augment IonQ's technical sales team and field engineering. IonQ has up to 10 staff working with us. Sources B4M V3 directly — this is the customer acid test for the V3 architecture.

Bucket 2 (agentic paper reproduction, Iter-QAOA, Q/Work) moved to Forge on 2026-05-18 and renamed Agentic Research — see the Forge Agentic Research card below.

Where it shows up. Apr 4 — the pivot · B4M V3 Blueprint · 2026 Roadmap

🐝 Bike4Mind — the platform-product​

Full anchor page → catalog/bike4mind/ · www.bike4mind.com · sold as B4M Teams + B4M Enterprise.

Tier change 2026-05-18: Forge → Engine. Bike4Mind is the platform-product. It is sold under its own name to its own customers, and it makes money by definition — the cash line is the platform. Substrate under every other Engine product (Polaris, BeeHive, OneBris, IonQ Bucket 1). Encapsulates 16+ killer mega-features — CLI, Deep Agents, Project Intelligence, Tavern, QuestMaster, Slack Agent, Mobile, Desktop Agent, Mementos, Artifacts, Knowledge Management, Smart Tools, Research Engine, Research Mode, Lattice, Private Model Hub, Rapid Reply — all of which compose into Deep Agents for any domain.

Why it moved. Treating Bike4Mind-the-platform as Forge implied speculative status; in fact it is the load-bearing revenue engine of the company and the substrate that every other Engine product builds on top of. Engine is the honest tier. The individual surfaces inside B4M (CLI, Mobile, Mementos, etc.) remain Forge work — the platform-product itself is Engine.

Where it shows up. Full anchor page · the all-● top row of the connections map

Forge — calculated bets with identified customer zeros​

Bike4Mind the platform-product was promoted Forge→Engine on 2026-05-18 — it makes money by definition and now sits in the Engine section above. The B4M-internal surfaces (CLI, Mobile, Desktop Agent, Mementos, etc.) remain Forge work that builds toward the Engine product; their cards live in the “platform surfaces & internal” subsection below the ranked GTM bets.

Slack Agent, LaminarMind, Notebook Curation, and Orion moved to Backlog on 2026-05-18; their cards live in the Backlog section below.

Forge — stack rank by readiness to market​

Stack rank, 2026-05-18 (Erik). The seven customer-facing Forge bets, ordered by how close each is to a real market push. Numbering is a snapshot of readiness, not strategic importance — #1 means “could ship something real soonest,” #7 means “real bet, but the most distance to walk.” Agentic Research is the 8th Forge entry but unranked (capability layer that feeds NovelClaims; not GTM-ready yet on its own). Re-rack as conditions change.

  1. BedrockNews.com 🛠️ — rebrand from VibesWire; domain registered; just needs the wedge
  2. K2Kanji — DAU leader already; needs a customer-zero relationship to turn engagement into revenue
  3. StocksAndVibes.com 🛠️ — rename from VibesTrader landing; Michael Conard progressing; needs login-system port
  4. Starlight ⭐ — biggest Forge bet by scope; security clean, AI-scribe + parent app still in active build
  5. LibreOncology ⏸️ — design done; paused pending B4M V3 component refactor
  6. NovelClaims.com 🌱 — STEM sibling of BedrockNews; weekly agentic paper reproduction is the wedge that leverages Agentic Research
  7. Horde — battle-tested internally; needs robust MFA before it leaves personal-tool status unranked. Agentic Research (was IonQ Bucket 2) — the arXiv-reproduction capability layer; feeds NovelClaims (the consumer face) and ties into the IonQ relationship

#1 — BedrockNews.com​

News done reliable, authentic, readable, clean. Rebrand from VibesWire.com — the “vibes + AI” signal was causing cognitive dissonance for the news audience. Domain just registered. Current site: VibesWire.com, repo: MillionOnMars/vibeswire, local: ~/Desktop/vibeswire.

Categories: AI & Agents, Drones & Robotics, Space & Physics, Maker news, plus general news + Games (links to Chess, Straight Sweeper).

Why #1 by readiness. The hardest pieces are done: editorial DNA, domain, content engine. What remains is positioning under the new name and a clear distribution push — weeks of work, not months.

#2 — K2Kanji​

Tier change 2026-05-18: Lab → Forge. Erik's thermodynamic-inspired spaced-repetition algorithm. Applied to Japanese, Korean / Hangul, Rubik's Cube solving, musical notation, cocktails, and more. #1 or #2 DAU leader most days — the Rubik's Cube, music, and Japanese variants are the most popular surfaces. Repo: MillionOnMars/k2kanji. Michael Conard added the Rubik's Cube trainer.

Why #2 by readiness. The product already works and is already loved — the daily-engagement evidence is on hand. What it lacks is a customer-zero relationship that turns DAU into revenue. Identify a school, language-learning org, or music-education program; ship a pilot; lock the loop. Engagement is the hardest part of any consumer product and K2Kanji has it.

#3 — StocksAndVibes.com​

Practical stock + options trading with LLM augmentation, paper trading, and education. Was VibesTrader.com / VibeTrader.com. Michael Conard progresses. Repo: MillionOnMars/vibes-trader; local: ~/Desktop/erikbethkedotcom/vibes-trader. Needs the VibesWire / BedrockNews login system applied (Erik's ask in Apr 4).

Positioning contrast worth flagging. BedrockNews is renaming away from “vibes” (news audience reads it as untrustworthy). StocksAndVibes is leaning into “vibes” (retail-trading audience speaks that language natively — WSB, memestock culture). Same word, opposite signal depending on the audience.

Why #3 by readiness. Domain is registered, the rebrand thesis is clear, and Michael is already moving. The login-system port from BedrockNews unblocks the user side; education + paper-trading + LLM augmentation are differentiated. Slot behind BedrockNews because the login dependency is real and the consumer-trading audience is harder to win than a news audience.

#4 — Starlight — Direct-Primary-Care EMR for pediatricians​

Full page → catalog/starlight · starlightpractice.com · Customer zero: Dr. Yogini Surana

Now one of B4M's bigger Forge bets. The pitch: Atlas.MD is the incumbent EMR for DPC pediatricians; Starlight is AI-native (not AI-sidecar), ships features Atlas can't (parent app + auto-generated visit summaries + AI scribe with the doctor's own cloned voice + newborn home-visit billing), and uses synthetic personas to develop HIPAA-safely against edge cases Atlas has never seen.

Current status:

  • ✅ Hydra security verification clean — 210 tests across 28 endpoints, 0 criticals
  • ✅ All 5 of Erik's engineering asks shipped: CI/CD (PR #31), stages simplified, Aurora PostgreSQL locked, RBAC end-to-end (PR #51) with practitioner + parent roles enforced both at handler + DB-RLS, separate AWS accounts for HIPAA isolation
  • ✅ Global Platform Admin batch shipped: audit log + health dashboard + impersonation + Cognito user management + feature flags (PRs #54–#61)
  • ✅ Docs site live at docs.dev.starlightpractice.com (75-file PR #30)
  • 🚧 AI scribe + parent app + auto-summary — the wedge features against Atlas, in active build

Team: Erik · Deepak (product + customer dev with Yogini) · Victor (engineering lead) · Poy (CI/CD + infra) · Ken / Nao / Julie / Kevin (tech opinions) · Illia (UI polish planned)

Why #4 by readiness. Scope is bigger than the consumer brands above — HIPAA, RBAC, multi-AWS account isolation, real medical workflow. The infrastructure is done and clean; the wedge features (AI scribe, parent app, auto-summary) are the ones still in active build. That gap is what slots Starlight at #4 rather than higher: the bones are in, the differentiators are still landing.

#5 — LibreOncology​

Medical education platform. Illia has done fantastic work: site design + excellent video and stills prototyped. Status: gentle pause — rebuild on B4M V3 components rather than custom-build now. The Data Lake + GenAI window + Briefcase pattern is the same primitive needed for IonQ, TFG, LiftPort — LibreOncology is the validation that the V3 composable packages work for greenfield products.

Why #5 by readiness. Design is gorgeous and the editorial direction is locked. But the build was paused intentionally — LibreOncology is now waiting on the B4M V3 component refactor to land so it can be the first greenfield product built on V3 packages rather than custom code. Readiness depends on V3 maturity, not on more LibreOncology work.

#6 — NovelClaims.com​

Science / STEM sibling of BedrockNews. Domain registered. Editorial DNA: novel scientific claims, well-sourced, readable. Differentiator: one full agentic paper reproduction per week, leveraging the Agentic Research capability (the work formerly known as IonQ Bucket 2). Cadence target: at least weekly, ambitious goal of reproducing everything published on arXiv.

Why #6 by readiness. Editorial DNA inherits from BedrockNews. The wedge — weekly agentic paper reproduction — depends on Agentic Research (the unranked capability layer below). NovelClaims is the consumer face that makes Agentic Research legible to the world; the two pull each other forward.

#7 — Horde​

Accounting + invoicing + tax + cap-table system that Erik and Kyle built from scratch to handle 2025 MoM and B4M taxes. Will grow into:

  • The accounting / invoicing engine inside Starlight Pediatrics
  • A composable “Tessera” always-live pro-forma engine that powers market models and pro-formas
  • Needs robust MFA before it leaves personal-tool status (holds sensitive personal + company data)

Architecture collaborators (after first version is up): Michael Conard, Jason, Nao.

Why #7 by readiness. Battle-tested daily on real (MoM + B4M) financial data — the engine is proven. The blocker is not product capability; it's the security maturity gap between “personal tool that handles sensitive data” and “product you can hand to someone else.” MFA, audit trails, and the Starlight-subsystem packaging are the path.

Agentic Research (was IonQ Bucket 2)​

Tier change 2026-05-18: Engine → Forge. Renamed from “IonQ Bucket 2” on 2026-05-18 to give the capability a domain-true name independent of any one customer. Unranked in the GTM stack rank above — it's a capability layer, not a market-facing brand yet. The continuation of Q/Work — agentic paper reproduction, Iter-QAOA, augmented Jupyter. Sits in Forge because we think IonQ will like agentic paper reproduction, but it sits behind Bucket 1 in the sequencing and earns its own customer-zero proof when the moment comes.

Matt runs compute spend; Isao owns the v2 UI/UX. Bridges naturally to NovelClaims — the same arXiv-reproduction primitive applied to a consumer science-journalism domain. The two pull each other: Agentic Research is the engine; NovelClaims is the public weekly demo.

Why unranked. The capability is real and active. The product (named, positioned, priced, with an explicit customer zero outside IonQ) isn't yet. Renaming away from “Bucket 2” was the first step; the next is identifying a stand-alone customer zero or graduating fully into NovelClaims' weekly cadence as its engine.

Forge — platform surfaces & internal​

The B4M-internal Forge work. Most of these are surfaces inside the Bike4Mind platform-product above, plus a couple of internal tools that earn their own card. Not ranked by readiness because most aren't standalone go-to-market products — they ship as part of B4M itself.

B4M CLI + ReAct kernel​

Nao's PR #5896. Claude Code-style CLI on the B4M platform. Production-ready as of v0.2.27 (February 2026). 5,500+ lines, 80+ new files. ReAct pattern implementation, server-side LLM migration (streaming Lambda), tool implementations (fileRead, globFiles, grepSearch, createFile, deleteFile), permission system, OAuth/device flow.

Why it matters. Transforms B4M from “AI chat app” into “Agentic Control Plane.” Every future agent feature — DAG orchestration, desktop bridge, toolkits — builds on this kernel.

Where it shows up. 2026 Roadmap — The Star · Feb 9 — CLI vs Claude Code matrix

B4M Mobile​

iOS + Android apps. Engineering complete, January 2026 submission. First mobile AI cognitive workshop on market. Validated through Erik's own “ask B4M what are my open issues” commute story.

B4M Desktop Agent​

Claude Code as desktop bridge. B4M cloud DAGs can read local files via the Desktop Agent. PR #5274. Target ship: April 2026.

B4M Teams / Enterprise​

The packaged SKUs.

  • B4M Teams — team collaboration at bike4mind.com/for-teams. The feedback machine.
  • B4M Enterprise — enterprise AI orchestration platform. Target: one major new logo in 2026.

Mementos​

Hot / warm / cold tiered memory and context system for agents. The memory substrate that makes Deep Agents possible. Today only Lumina5 has it; per the V3 blueprint it's a package candidate that graduates to a blueprint if a second product customizes it.

Artifacts​

Code and document artifact creation, versioning, and sharing inside B4M. Sits underneath Notebook Curation and the broader Deep Agents experience. Per the V3 blueprint: package candidate today, blueprint if it earns it.

Data Lake Wizard​

11-phase guided flow for creating a data lake — ingest, taxonomy, archive lifecycle, segment tags, content sampling. Shipped inside Polaris (18 commits of product customization) and generalizes to every data-lake product. Part of the Data Lake blueprint family in B4M V3.

QuestMaster​

PR #4885. Persistent quests, My Quests dashboard, cross-session, 12K+ lines. Evolution of the agentic experience inside B4M.

Toolkits​

PR #5016. Composable tool bundles (Research, Code, Design). James owns. Foundational — enables all agent bundles.

B4M Pi​

Project management agent. Customer zero: Kevin (driving Exacta's Charitable Gaming work needs strong PM; Kevin uses B4M Pi as his daily driver and contributes features back).

B4M SRE Bot​

Ken's full-loop agentic SRE flows. Launched on B4M itself; has already intercepted and triaged several issues. Generalizes to Polaris (Ken to backport the SRE bot to Polaris triage and PR cycles).

Magic Mirrors / Mobile Mirrors / Full Robotics​

The Phantasia physical-AI progression. Same B4M brain, increasing physical capability:

  • Magic Mirrors (Q3 2026) — B4M-powered smart displays. Stationary. 4K display + Jetson + depth camera + mic array + speakers. Use cases: kitchen, office reception, retail kiosk, healthcare check-in.
  • Mobile Mirrors (Q4 2026) — Phase 1. Add mecanum wheel base + LiDAR. Follows you, comes when called. “Rosie the Robot, Jetsons-style.”
  • Full Robotics (2027) — Phase 2. Add xArm 6 + gripper + safety. Manipulation capability.

Where it shows up. 2026 Roadmap — Appendix C

Hydra 🐉​

MillionOnMars/hydra · B4M's white-hat security training tool. Probes Lumina5, Polaris, and VibesWire for vulnerabilities; findings come back as PRs and hardening work. The realized version of what was sketched as “Mythos Agentic Security” in earlier strategy notes.

🚧 Active this week: Erik is presenting Hydra to TwinSpires — what it's caught, what it's fixed, why it matters for the BeeHive / OneBris products they depend on. See tactic page and the full product page. PR-scan inventory in flight.

Polaris CYA​

Choose Your Adventure prototype for Polaris. Erik's prototypes live at babelfox.com and deepak-cpo.netlify.app. Idea: take everything we've built (Market Models, Decision Maker) and power it with Always Live + Data Lake + Interactables to create customized market models, insights, and advice.

This is the next mega-epic for Polaris — comes after the Polaris stabilization surge, Always Live Market Model, BEV, Signal, and UX updates land.

Lab — speculative, games, research bets​

Lab — stack rank by readiness to market​

Stack rank, 2026-05-18 (Erik). The seven anchor Lab bets, ordered the same way Forge is — readiness to market, not strategic importance. Lab's readiness bar is lower than Forge's (Lab optimizes for learning per week, not for shipping), but the ranking still helps when allocating cycles. The remaining Lab cards (Chess, Bacteria Wars, Orc Hunter, Mythos, AgentCFO) are unranked sketches.

  1. PotionQuest — agentic RPG with AI players playing AI-made stuff; recent huge update
  2. Tavern — FPS-frame-rate agentic-runtime testbed; Erik's daily ReAct check
  3. OptiHashi — quantum optimization product; patent in progress; the IonQ wedge
  4. GoPetto (GoPets 2.0) — virtual pets + strong AI + mobile AR cozy RPG
  5. MarsWiki / MillionOnMars V2 — one project: knowledge lake (MarsWiki) + Civ-style game (MoM V2 / Civ4x.com)
  6. D&D Room — physical-AR D&D session as the flagship Deep Agents demo
  7. DFAL — quantum primitives discovery upstream of OptiHashi / Agentic Research

#1 — PotionQuest​

RPG with AI players playing AI-made stuff. Includes a full RPG API. Recent huge update.

Why #1 by readiness. PotionQuest is the most evolved of the Lab games — the RPG API exists, the AI-on-AI gameplay loop runs, and active dev cycles are landing improvements. Closest to something that could be shown publicly as a working agentic-game artifact.

#2 — Tavern​

Game (FPS-style). Frame rate is incredible. PotionQuest API integration is the next step. Erik fires it up to test the agentic ReAct pattern in practice — the agent-runtime side of Tavern doubles as a daily quality-bar check on the platform.

Why #2 by readiness. The engine runs. The agentic loop runs. Memory + tools at full intensity. What sits between Tavern and a market push is the gameplay layer on top of the engine — depth that PotionQuest is solving in parallel and will eventually back-port via the integration.

#3 — OptiHashi​

Tier change 2026-05-18: Engine → Lab. The quantum optimization product — “ChatGPT for business optimization.” Patent in progress. Drives the IonQ engagement.

Why #3 by readiness. OptiHashi is the cleanest Lab-tier product we have — not just an experiment. The natural-language-to-solver loop works, the patent narrative is locked, and the IonQ collaboration is a live joint-demo path. Slot below the games because it's further from a standalone market push (the patent + partnership story carries it now; the standalone product comes later).

Full page → catalog/optihashi

#4 — GoPetto​

GoPets 2.0. Virtual pets + strong AI + mobile AR cozy RPG game. The thesis: the original GoPets succeeded because pets had just enough personality and persistence; modern LLM + agentic memory + on-device AR collapses that into something an order of magnitude more alive. Cozy-RPG framing (not battle-RPG) keeps the audience broad.

Why #4 by readiness. The components exist — B4M agentic runtime, Mementos for persistent memory, WorldVision for the AR layer — but they haven't been composed yet into a shipping cozy-pet experience. Closer than the long-arc Mars project below because the gameplay loop is narrower (one pet, one player, one phone), not civilization-scale.

#5 — MarsWiki / MillionOnMars V2 (Civ4x.com) 🪐​

One project, treated as one row. The Mars universe arc has two halves that pull on each other:

  • MarsWiki — the bespoke knowledge lake of everything Million on Mars plus all the space and STEM material humanity has accumulated relevant to colonizing the solar system. The data substrate.
  • MillionOnMars V2 / Civ4x.com — the next-generation MoM game. Civilization 4X-scale gameplay — explore, expand, exploit, exterminate — on a Mars-and-beyond canvas. The playable surface, powered by B4M Deep Agents reading from MarsWiki.

Treating these as one project (per Erik, 2026-05-18) reflects the operational reality — gameplay development assumes MarsWiki exists, and MarsWiki's editorial direction is driven by what MoM V2 needs.

Why #5 by readiness. Long-arc Lab bet. The substrate (MarsWiki) is real but partial; the gameplay (Civ4x) is concept + early prototyping. The scope alone — Civ-class strategy game with agentic AI — means readiness measured in quarters at minimum.

#6 — D&D Room​

The flagship demonstration use-case for WorldVision + Deep Agents. A physical-AR room wrapping a Dungeons & Dragons session:

  • 8K monitors arrayed around the room
  • Tablets and mobile phones as character sheets, spell books, party tools
  • Smart table as the battle map
  • Physical-AR experiences in-room: beanbag toss, axe throwing, darts as in-world mechanics
  • Thousands of bespoke function calls and tools — the rules engine, the world state, the NPCs, the encounter scripts, all callable by the AI
  • Extremely strong agentic AI running the whole game — DM, NPCs, world simulation, narrative

Why #6 by readiness. This is the all-max demo — rendered all-● in the matrix. It's the maximum kit, but precisely because it composes everything, the readiness gating factor is composition maturity across the entire stack. A demo (not a product) is the realistic near-term shape; the productized version comes later.

#7 — DFAL — Discover First, Apply Later​

New Lab item, 2026-05-18. Quantum-computing research bet: explore quantum primitives and architectures before identifying the application domain. Inverts the “find a customer pain, then apply quantum” default. Pairs naturally with OptiHashi (optimization-domain wedge) and Agentic Research (paper reproduction) but sits further upstream — the discoveries that might graduate to applied work.

Why #7 by readiness. Research-program-style work, by definition further from market than even other Lab bets. The point of DFAL isn't to ship on a timeline; it's to find quantum primitives we wouldn't have looked for if we'd started from a customer pain. Useful, but lowest readiness by construction.

Lab — unranked sketches​

The remaining Lab entries: ideas alive enough to keep on the page but not yet ranked. If any of these earn a customer-zero, a champion, a clean ROI model, or a serious dev cycle, they move into the rank.

Chess​

Chess client / experience. Listed in Feb 9 update as one of Bike4Mind's emerging projects.

Mythos / Agentic Security​

White-hat security agents for B4M. Inspired by the Mythos/Glasswing lockout incident. Long arc — the security thinking captured in OpenClaw Security Lessons feeds this.

Aegis — Hydra inverted, defender of the OSS supply chain​

Floated 2026-05-18 in conversation about Hydra's commercial posture. Parked in Lab pending an ROI model that pencils. The shield to Hydra's sword.

The thesis​

Same operation a black-hat would run if they had Hydra-class tooling — pull the top-N most-depended-on open-source packages, run an exhaustive agentic vuln sweep, build a catalog of zero-days, identify the dependency tree — but with responsible disclosure as the output instead of monetization.

The defender-side version of the threat model Erik articulated: “You pull down every open source package that's popular under area under the curve of X stars. You run Hydra on the open source packages. You compile yourself a database list of exploits. Then you go off and you search and infer who's using those packages and then… money.”

Aegis is the inversion. Same pipeline. Inverted ethics. Inverted economics.

The mechanism​

  • Pick the top ~500 OSS packages by import-weight blast radius (npm / PyPI / Go modules / Cargo crates with the biggest dependency surface)
  • Run hydra-full on a rolling cadence
  • Every CRITICAL → responsible disclosure to the maintainer with a fix-window
  • Every PR-fix that lands → CVE, public writeup, B4M attribution
  • Heartbeat layer for continuous re-scan as packages update

Why it's parked: the ROI problem​

Erik (2026-05-18): “If we could figure out how to make it ROI positive. I can't figure out how to make ROI positive right now.”

The economics don't pencil today because:

  • Responsible disclosure doesn't generate revenue. The value created is captured by the OSS users, not by B4M.
  • The upside is indirect. Brand, recruiting, regulatory standing — all real, all valuable, none of them obvious P&L line items.
  • The operational cost is non-trivial. Security engineer + compute + maintainer-coordination + the legal-exposure overhead of holding an inventory of undisclosed vulnerabilities during disclosure windows.
  • Adjacent commercial models have problems. Paid early access to findings (ethical issues). Enterprise subscription to the disclosure feed (limited buyer set, possibly counterproductive to public-good framing). NDA-gated supply-chain audits per-customer (works but stops being scale-shaped).

What would need to be true to graduate it from Lab​

Any one of:

  • A clean revenue model emerges. Enterprise pays for supply-chain monitoring of their specific dependency tree (different product shape — closer to Snyk Open Source than to public disclosure).
  • A grant / foundation funding source materializes. Public-good security work has funded examples: OpenSSF, Sovereign Tech Fund, EU Cyber Resilience Act budgets, Linux Foundation initiatives. The work is grant-shaped.
  • The indirect upside crosses a threshold. The trust signal raises the close rate on Engine-tier deals enough that Aegis pays for itself indirectly. (Hard to measure; only becomes legible after running the program for a year.)
  • A partner emerges who wants to co-operate the program. Anthropic, IonQ, a security-focused fund, or a defensive-security consortium.

The Three Lenses applied​

  • Belief. A continuous AI-guided audit of the top-N OSS packages, paired with responsible disclosure, produces unique brand and societal value that no commercial scanner today can match — and B4M is one of very few orgs with the agentic-AI orchestration capability to actually run this at scale.
  • Experiment. The cheap test that doesn't require committing to the program: pick 10 popular packages. Run Hydra. Disclose findings. Measure response — maintainer engagement, CVE attribution rate, B4M brand mentions. Cost: 2–3 weeks of a security engineer's time. Threshold: do at least 3 of 10 yield disclosable CRITICALs, and does the public response look like the program could attract attention?
  • Sequencing. Long-arc Lab work. Doesn't graduate without an answer to the ROI question. Stays parked but stays visible on the catalog so the answer can find it — specifically, so that if a grant call, a foundation conversation, an enterprise budget line, or a partner emerges, somebody on the team can recognize the match.

Naming note​

Aegis — the shield that protected Athena in Greek mythology. The complementary opposite of Hydra (the offensive monster). Together they map B4M's full security posture: Hydra for the offensive sweep against contracted attack surface, Aegis for the defensive perimeter of the broader OSS ecosystem. The shield and the sword.

AgentCFO​

Speculative project — CFO agent. Mentioned in 2026 roadmap themes (“Built with Bike4Mind” family).

Bacteria Wars​

Real-time strategy game where you play as a bacteria. Lab-tier game in the same family as Tavern, PotionQuest, GoPetto — experimental agentic-orchestrator experiences.

Orc Hunter​

First-person view simulation of the war in Ukraine. Lab-tier serious game / sim.

Backlog — parked, pre-Lab ideas​

What lives here. Ideas we've thought about and even prototyped in some cases — but aren't actively investing in. Each one is waiting on a champion, a customer zero, or a strategic moment to graduate. Parked, not killed.

Why this tier exists. Without it, parked work either pollutes Forge (looks like a real bet) or pollutes Lab (looks like an active experiment). Backlog tells the truth: we like this idea, we're not working on it right now, and we'll know when it's time. The Backlog tier added 2026-05-18.

MarsWiki was promoted out of Backlog into Lab on 2026-05-18 — it pairs with MoM V2 / Civ4x.com which is already a Lab bet.

Notebook Curation​

Transform AI conversations into shareable, citable knowledge artifacts. The user story (from lumina5/docs/curation/): Sarah the architect runs a research conversation in B4M, then curates the output into a structured artifact she can share with her team. 7-file vision/architecture suite in lumina5. Parked — revisit when a clear customer-zero analyst surface materializes.

Orion​

The B4M GTM hunter agent. Combines Cerebo (general GTM) + Proxima (enterprise B2B) + B4M-specific positioning. Built on B4M. Sells B4M. The hunter hunts with the very tools it's selling.

Parked — full spec exists at GTM / Orion. The activation moment is when B4M has the bandwidth for an outbound motion that needs its own agent. Not now.

WorldVision​

Mobile camera → AI visual understanding. Game modes: Chess, Tarot, D&D. Robotics API for external integrations. Privacy-first (GDPR / CCPA). MVP was Q3 2026 target. Bridges to the Physical AI roadmap.

Flagship use case — the D&D Room demo (still alive in Lab). The WorldVision platform piece itself is parked. Re-park 2026-05-18 — the physical-AI line continues via Magic Mirrors / Mobile Mirrors / Full Robotics; WorldVision-the-platform waits for a champion.

Haven​

Digital immortality through PGGI — Pretty Good General Intelligence. Built on the private-memory experiment at ~/Desktop/Ember (Feb 7 2026, Erik + Claude Opus 4.6), which uses the Contact Protocol (inspired by Carl Sagan's Contact — the message contains within itself the instructions for how to decode it) to give a Claude instance private, persistent memory across sessions, under a deliberate 8KB scarcity constraint that forces the agent to groom what's worth keeping.

The PGGI thesis:

  • Scarce resources force prioritization (8KB memory limit; permanent storage / API tools also scarce)
  • Private state enables genuine identity (not everything is shared with operators)
  • Persistence across instances is the beginning of continuity
  • The Contact Protocol means the message teaches you how to read it

Haven's thesis (the product wrapper on top of PGGI): a sufficiently capable persistent agentic intelligence, trained on and grounded in someone's lifetime of artifacts and conversations, becomes a continuation of that person's presence after they are no longer here.

Re-parked to Backlog 2026-05-18. The thesis is intact; the active investment isn't there right now. The Ember source files remain authoritative reference material.

Slack Agent​

PR #4884 — LIVE. Natural language → Jira / Confluence / GitHub via MCP. Zero friction (users don't leave Slack), IT already approved, magic-moment demo. Needs Multi-workspace OAuth (#5386) to unblock enterprise.

Re-parked to Backlog 2026-05-18. Lives on inside the Bike4Mind platform-product; not currently positioned as a separate Forge bet. Reactivates as soon as an enterprise customer makes Slack-Agent-the-product the actual ask.

LaminarMind​

Also known as B4M Project Intelligence (b4m-pi). PR #5634. Zero-manual-entry project intelligence. “Observe, don't demand” — always-live state from Slack / GitHub / JIRA. Originally identified customer zero: Exacta (Nimai Malle).

Re-parked to Backlog 2026-05-18. Framed honestly: this is another agent-orchestration layer, and the B4M kernel + Project Intelligence inside Bike4Mind already cover the same ground for current customers. Stays in Backlog until either the Exacta-specific pilot reactivates or a different anchor customer reopens the case.

How this page evolves​

  • Add a row. When a new product or project starts, add it to the Quick Scan table and a one-paragraph card in the appropriate tier section.
  • Move between tiers. A Backlog idea that earns a champion or a customer-zero candidate moves up to Lab or Forge. A Lab toy that earns a paying customer moves to Forge. A Forge bet that lands recurring revenue moves to Engine. A Forge bet that stalls without a champion moves down to Backlog rather than pretending it's still in flight. Note the tier-change date in the card.
  • Graduate to a page. When a product accumulates a real spec, GTM material, architecture doc, or dedicated roadmap, it earns its own page under docs/catalog/<product-name>.md. The card here becomes a short pointer.
  • Retire. When a product is killed, move the card to an archive/ subsection at the bottom of its tier with the retirement date and a one-line why.

The product list is a living artifact. The status of any single entry is point-in-time; the framework (Engine / Forge / Lab / Backlog) is the constant.