Live brain
0
brain nodes
+0
today
+0
this week
Top: articles (45) · decisions (32) · facts (28)
Quality gate
Last 50 reviewer decisions:
86% accepted first try
12% accepted on revision
2% flagged for human
Workforce reviews itself. Catches its own mistakes.
The math
Most builders pay for a stack of separate AI tools. Keen routes across 12+ models in one subscription — chat, code, build, and workforce in a single seat.
With Keen OS
€49/mo
Cancel anytime · 1,000 SPARKS included
What you get
Chat · smart-routed across 12+ models
Build · idea → live preview → deployed
Code · shared memory across every surface
Workforce · recurring agents with approval gates
Keenoble is designed to consolidate AI tools for thinking, writing, coding, and building. It does not replace image generation (Midjourney), music (Suno), or video (Runway). Competitor prices from public pricing pages, May 2026.
Room 01 · Chat
Other AI chats use one model for everything. We route every prompt to the model best-suited for the task — in milliseconds. Speed for casual questions. Depth for reasoning. Specialists for code, long context, and verification.
Groq Llama 3.3 · speed
Routed for quick answers and casual queries where latency matters most.
Room 02 · Build
No copy-paste. No re-explaining. The build inherits everything from your conversation — the goal, the constraints, the half-thoughts. You ship, edit and iterate without leaving the flow.
CHAT
I want a pricing page for a SaaS,context captured
BUILD
Live preview readycontext inherited · 0 prompts
SHIPPED
your-pricing.keenlabs.applive · public
When to reach for build
When chat sketched the idea and you want it live
When you want to ship a landing page, app, or prototype
When you don't want to think about deploy infrastructure
Room 03 · Code
Your chat. Your build. Your code. One memory layer. One context graph. Edit a paragraph in chat, and the code knows. Refactor a function in code, and the build updates. We've stitched the toolchain into one nervous system.
CHAT
Your conversation
BUILD
Visual preview
CODE
Source of truth
Encoded in shared memory
Project intent
Style decisions
User constraints
Past edits
Acceptance criteria
Test outcomes
“Other tools give you Chat. Or Build. Or Code. We give you the connections between them. The memory. The continuity. The thing that makes solo founders move like teams.”
When to reach for code
When AI got 90% right and you want surgical precision
When you're integrating with existing repositories
When you want full IDE control with shared context
Room 04 · Speed & Smart Routing
Other platforms use one model for everything — you pay frontier prices even for simple questions. We route across 12+ models: frontier when the task demands it, fast and cheap when it doesn't. The router decides in milliseconds.
The moat · Smart routing
Claude Opus 4.7
deep reasoning
GPT-5
synthesis
Groq Llama 3.3
speed
DeepSeek V3
code
Kimi K2
long context
Gemini 2.5 Pro
general
Each model is matched to what it does best. The router picks in milliseconds based on prompt characteristics — you never manually choose.
Cost efficiency
Pay for what you need
Frontier models for deep reasoning. Fast models for quick answers. You never overpay on simple tasks.
Design principle · no independent benchmark
Right-sized latency
Fast when fast matters
Sub-second for casual queries. Deliberate for complex synthesis. The router matches speed to task complexity.
Design principle · no independent benchmark
Model intelligence
Right model, every time
Code goes to code specialists. Long documents to long-context models. No prompt hits a model that wasn't built for it.
Design intent
Access
Chat, Build, Code — with smart routing, shared memory, and verified output. €49/month. Cancel anytime. Currency: SPARKS ⚡ (1,000 per month).
Or scroll on for the Workforce tier ↓
Commercial Product Layer
The live preview is the operating map, not a side project. Keenoble at ai.keenlabs.pro is the mainstream commercial product; Keen OS Workforce is the custom orchestration layer being built from the same chat, build, code, memory, verification, and ontology foundation.
01 · Start
€49/month
Mainstream Keenoble plan with 1,000 Sparks credits designed to cover normal monthly chat, build, and code usage through smart model routing.
1,000 Sparks credits
smart chat model routing
build preview mode
code precision mode
02 · Scale
€180/month
Higher-context Keenoble plan designed for 5x usage, project orchestration, multi-project context, and a verified project pipeline.
5x Premium usage
project orchestration
full context over multiple projects
high-context personal project agent
03 · Orchestrate
from €2500
Custom recurring skill-workforce orchestration for source-watch loops, eval loops, PR loops, verification gates, and human approval workflows.
custom automated skill workforce
recurring missions
source-watch and eval loops
PR loops with human approval gates
How Keenoble works
Chat routing layer
Routes everyday questions and deeper work to suitable model paths instead of forcing every prompt through one model.
Build preview layer
Lovable-like visual preview direction for seeing what is being built while product work takes shape.
Code precision layer
Code mode for precise surgical code edits, implementation work, and reviewable changes.
Project orchestration layer
Pro direction for multi-project context, a verified project pipeline, and a high-context personal project agent that tracks updates.
Workforce automation layer
Custom mission layer for recurring source-watch, eval, PR, verification, and human approval loops.
Memory and Ontology
Keen Labs is moving toward a four-layer memory system: working memory, project memory, source memory, and mission memory.
Felix is the internal memory and coordination concept used to describe the project-agent layer that tracks updates and preserves context.
Obsidian-style Brainnodes are being developed as the visual knowledge graph and ontology interface for project context.
Ontology-backed project context connects actors, concepts, cases, agreements, sources, agents, and missions before customer-impacting output is approved.
Proof-of-work systems
Pensionskollen
Proof-of-work system for regulated source-of-truth guidance and workforce-backed verification loops.
LöntagarPension
Proof-of-work surface for employee-focused pension, salary, insurance, and benefits context using the same workforce pattern.
Customer-impacting publishing, merging, and release decisions are designed to pass through verification and human approval gates unless explicitly enabled by the customer.
Case Study
PENSIONSKOLLEN
Active since May 2026. Swedish pension and payroll guidance platform. Source-backed advisory content for workers, freelancers and HR professionals navigating Sweden's retirement system.
We attached a Keen OS workforce to Pensionskollen at launch. The workforce drafts articles, monitors source and ranking signals, and verifies claims against primary sources — the Truth Auditor blocks anything it can't cite before human approval.
First proof-of-work system for the workforce pattern. Zero paying customers yet — the goal is to prove the model works end-to-end before charging anyone.
→ pensionskollen.comPart Two
Premium is the daily driver. This is what's coming next: a human-approved workforce system for recurring missions, source watches, verification loops, and PR workflows. Not generally available as a self-serve product. Yet.
PROOF · LAST 24 HOURS
Two agent-drafted PRs merged after review. Twelve brain notes auto-written. Truth Auditor caught its first hallucination and refused to save it. Everything below is real — pulled from git and Supabase, not staged.
0
reviewed PRs
0+
brain notes
0
agents online
0+
routed models
0
sources cited
0
sprints shipped
What the workforce did while you slept.
Nine sprints. One week.
SPRINT A
Cleanup + harden
Two agent-drafted PRs as proof.
shippedSPRINT B
Felix-lagret
Four-layer brain. Skills library.
shippedSPRINT C
Truth Auditor
Verifier-gating on every output.
shippedSPRINT E2
Stripe + nav
No more lost payments.
shippedSPRINT E3
Editing fixes
Visual edit. Stream recovery. Chat edit.
shippedSPRINT W1
Pensionskollen
First project workforce-attached.
shippedSPRINT K6
keenlabs.pro v3
Six new sections. This page.
shippedSPRINT O1
Ontology
Typed graph. Sources. Workflows.
buildingSPRINT W2
Keen OS dogfood
Workforce on itself.
nextA workforce that never forgets, never bullshits, and gets better while you sleep.
KEEN LABS · LANDSKRONA · SE · MMXXVI
System Updates · Static Preview
This section is static for now: it shows the latest shipped layers and the work currently being shaped into recurring Keen OS Workforce missions. It is ready to be wired to changelog and mission data later.
Operating relationship
Keenoble routes intelligence. Keen OS Workforce runs missions. Pensionskollen publishes verified guidance. LöntagarPension captures employee-focused search intent. keenlabs.pro shows the system being built.
Latest shipped layers
Routing layer for selecting intelligence paths across product, build, code, and verification work.
Safer structured-output path for agent responses that need to be parsed, reviewed, or stored.
Guardrails for Swedish pension, payroll, insurance, and benefits content where source quality matters.
Source gating that prefers official Swedish sources before regulated guidance is published.
Cases can carry visible metadata so readers and agents can understand the regulated context.
Actors, concepts, cases, agreements, and sources are represented as content graph inputs.
Content and mission outputs can expose verification state instead of hiding source confidence.
Evaluation matrix for search, generative answers, and AI-overview readiness without claiming outcomes as live traffic.
Now building
01
Defining repeatable mission specs before allowing recurring agents to run them.
02
Closing the loop from official source discovery to content update, reviewer check, and publish decision.
03
Connecting actors, concepts, cases, agreements, sources, agents, and missions as the planning graph.
04
Preparing scheduled checks that can identify changed source material and missing guidance opportunities.
05
Designing agent-generated PR workflows that still require a human approval gate before release.
Product ecosystem · Static map
This is not a portfolio wall. It is the current operating map: Keenoble routes intelligence, Keen OS Workforce runs missions, Pensionskollen publishes verified guidance, LöntagarPension captures employee-focused search intent, and WSP Screener is the market screening product queued for the same workflow.
Keenoble / ai.keenlabs.pro
Routes intelligence
Smart model routing AI engine and user platform that chooses model paths for chat, build, code, and verification work.
source: founder_self_reported · as of 2026-05-31
Open surface ↗Keen OS Workforce
Runs missions
Recurring agent mission layer for source-watch, content-gap detection, PR drafting, verification, and human-approved release loops.
source: founder_self_reported · as of 2026-05-31
Open surface ↗Pensionskollen
Publishes verified guidance
Swedish source-of-truth product for pension, payroll, and insurance guidance using regulated source guardrails.
source: founder_self_reported · as of 2026-05-31
Open surface ↗LöntagarPension
Captures employee search intent
Employee-focused authority satellite for salary, pension, insurance, and benefits questions.
source: founder_self_reported · as of 2026-05-31
Open surface ↗WSP Screener
Screens markets
Market screening and investing intelligence product queued for workforce-backed research loops.
source: founder_self_reported · as of 2026-05-31
Public URL pending
Room 05 · Memory
Every conversation, every decision, every truth audit feeds back into a shared knowledge graph. Agents don't start from scratch each task — they inherit the entire history. The workforce gets smarter every night.
347
brain entries this week
2
evergreen principles (immutable)
02:00
nightly self-rebuild (UTC)
Room 06 · The Ontology Layer
Keen Labs models regulated work as actors → concepts → cases → agreements → sources → agents → missions. Pensionskollen is the concrete example: Swedish institutions, benefit concepts, employment cases, collective agreements, official sources, verifying agents, and recurring missions are connected before output ships.
actors
concepts
cases
agreements
sources
agents
missions
Actors
Concepts
Cases
Agreements
Sources
Agents
Missions
CHIEF OF STAFF
Orchestrator
Splits goals into tasks, allocates agents, decides merge thresholds. The only agent that speaks directly to the operator.
MODEL
Claude Opus 4.6
ENGINEERING
Code agent
Writes, refactors, and debugs across multiple files with full repo context. Drafts PRs for review and approval.
MODEL
Claude Sonnet 4.6
TRUTH AUDITOR
Verifier
Every factual claim is verified against primary sources before it ships. Blocks low-confidence output. Non-negotiable gate.
MODEL
Perplexity Sonar
DEVOPS
Infrastructure agent
Owns CI pipelines, deploys, and infrastructure. Diagnoses build failures. Rolls back on degraded metrics.
MODEL
GPT-5
ARCHITECT
Design agent
Translates intent into technical specs. Designs APIs, schemas, and data flows. Writes the blueprint before Engineering builds.
MODEL
Claude Opus 4.6
MEMORY STEWARD
Knowledge agent
Aggregates daily session output into evergreen entries. Decides what the workforce remembers permanently. Runs at 02:00 UTC.
MODEL
Claude Haiku 4.5
Access · Workforce Tier
Keen OS Command is deployed in engagements, not sold on a pricing page. If you're building something real and want recurring agent systems with verification gates and human approval — tell us what you're working on.
Book 15 min →or join the waitlist
No sales funnel. No automated follow-up. Just a conversation.
The builder
Founder of Keen Labs (Q1 2026). Based in Landskrona, Sweden. Building AI systems for real-world workflows.
Previously: product and operations across fintech, payroll, and pension infrastructure. That's where the problems are. That's why Keen Labs exists.
The workforce running Pensionskollen is the same architecture available to Keen OS Command clients. It wasn't built to demo well. It was built to run.
