Polaris
The Product
A company operating system built for humans and AI agents
Polaris is an internal operating system for running a company. It replaces the patchwork of spreadsheets, Notion pages, Slack channels, and scattered markdown files with a single dashboard that holds goals, metrics, updates, the roadmap, and company values.
The thing that sets it apart is that both you and the AI read from the same place. The AI runs locally on your machine, and the project lives in a monorepo so the agent has full context.
Timeline
Feb 2026 – Present
Platform
Internal Web App
Role
Product Designer + Developer
Home dashboard — click to enlarge
The Problem
Running a small company means juggling a handful of tools that don't really talk to each other — and none of them were built for a world where AI is part of the team.
Tool Sprawl
Goals in a spreadsheet, updates in Slack, metrics in a dashboard, docs in Notion. Every context switch costs time and makes it harder to see how everything connects.
Fragmented Knowledge
Company knowledge is split across services. Connecting a goal to a metric to a roadmap initiative usually means stitching three tools together by hand.
AI Without Context
Most AI tools require you to paste context into a chat window. They don't have ongoing access to your goals, metrics, or company values, so the suggestions stay generic.
Cloud Lock-In
Most dashboards need accounts, subscriptions, and a server somewhere holding your data. For an early-stage company, that's more overhead than it's worth.
The idea behind Polaris is to put everything in one place that both you and an AI can read from, with the AI running locally via Ollama (or your own API key if you prefer). No cloud dependency, no API costs.
The Stack
Deliberately opinionated — zero cloud dependencies for the core product. Everything runs on your machine.
Next.js 16
App Router with React 19 and server components. Fast to build, easy to reason about, and the file-based routing maps cleanly to each dashboard tab.
SQLite
Zero-config, single-file database. No cloud account, no connection strings — your data lives on your machine as one file.
Tailwind CSS
Utility-first styling paired with JetBrains Mono for a monospace, retro-modern feel. Dark mode is the default.
Ollama + Gemma 3
Local AI inference via Ollama. The model runs on your machine, with no cloud API needed — so there's no API cost, and it works offline.
Vercel AI SDK
Streaming responses with a provider abstraction. Swap between local models (Ollama) or bring your own API key — the interface stays the same.
Monorepo
App code, documentation, brand guidelines, and agent instructions all live in one repository, so the development agent has the same context I do.
Core Features
Each tab in Polaris serves a specific purpose. I only added screens I'd actually use day to day.
Values
Company identity rendered as interactive cards covering brand voice, design language, color theory, and vision. The AI reads from the same source, so its output stays on-brand.
Updates
Async standups that replace daily meetings. Structured fields — working on, blocked by, shipped today — with an AI draft button that pre-populates based on recent activity. A summarize button digests all updates into a single overview.
Goals
Quarterly and yearly goal tracking with visual progress bars. Goals are seeded from the company's identity document, so there's always alignment between what you say matters and what you're tracking.
Metrics
Key business metrics with sparkline charts and trend indicators. Revenue, MRR, churn, activation rate — with support for custom metrics. Log entries with optional notes for context on why a number changed.
More Features
The rest of the dashboard — planning, documentation, and a persistent AI conversation.
Roadmap
Initiatives grouped by quarter with status tracking. Link roadmap items to goals, assign owners, and generate PRDs with a single click — the AI has full context on what you're building and why.
Docs
Renders all markdown files from the monorepo's docs folder. Specs, ideas, brand guidelines — accessible without leaving the dashboard or opening a separate editor.
AI Chat
A persistent AI conversation interface that runs locally. Ask questions about your goals, brainstorm roadmap ideas, or draft messages — the AI has access to everything in the dashboard.
Contextual AI Actions
AI that acts on your data, not just displays it.
Hover over any entity to surface contextual AI actions — click to enlarge
Hover over or select any entity in Polaris — a goal, an update, a metric, a roadmap item — and a floating action button appears with AI-powered actions specific to that context. Because the AI has access to everything else in the workspace, those actions can use the surrounding company state too.
Create Task
Turn any piece of content — a blocked item, a goal milestone, a metric anomaly — into an actionable task with one click.
Draft Message
AI composes messages with full context — it knows who you're writing to, what they're working on, and what matters to the company right now.
Generate PRD
Draft a product requirements document directly from a roadmap item. The AI pulls in related goals, metrics, and company values to create a grounded spec.
Summarize
Digest complex content — a week of updates, a long doc, multiple metric entries — into a concise overview you can act on.
Research
Find related context across your workspace. Blocked on something? The AI searches your goals, docs, and updates to surface relevant information.
Update Progress
AI-assisted goal progress updates. The AI suggests a progress percentage based on recent activity, shipped items, and metric trends.
Different entity types surface different actions — the menu adapts to what you're working on
Agent-First Development
How building inside a monorepo changes the relationship between you and your AI development partner.
On Kaizen Tracker, I used Claude Code as a development assistant — I'd describe what I wanted, it would scaffold components, and I'd review the output.
Polaris is different because the monorepo holds more than code. It also contains the company identity doc, brand guidelines, product specs, and architectural decisions. The agent doesn't need me to paste context into the chat first — it already has it. When it writes a component, it has the color palette; when it drafts a PRD, it has the company values; when it suggests a metric, it has the quarterly goals.
It also keeps related artifacts in sync. If I ask it to update a feature, it can update the docs and the marketing copy at the same time, so I don't have to remember to do that the next time I open those files.
Reflection
What building Polaris taught me about designing for AI-augmented workflows.
Opinionated Defaults
SQLite over Postgres, local AI over cloud APIs, a single repo over multiple. Each of those choices removed something I'd otherwise have to manage and made the system simpler to reason about.
Values in the App
Encoding company identity directly in the app as a rendered page (rather than a separate PDF) changed how I thought about it. The values became something the rest of the app could actually read from.
Two Audiences
Every information architecture decision had to work for both me and the AI. Structured data, consistent naming, predictable locations — the things that make an app easy for a person to navigate also make it easier for an agent to read.
Next Steps
- Multi-user team support — extending Polaris for small teams with shared goals and individual dashboards
- Mobile companion — quick updates and goal check-ins from your phone
- Plugin system — custom AI actions that teams can define for their specific workflows
- Webhook integrations — pull in external data (GitHub commits, revenue from Stripe) automatically
Thank you
Thank you for your time reviewing my work on Polaris!
If you'd like to get in touch, please say hi!
claytonanderson.work@gmail.com