PITH v0.1Open app →
The missing layer between thinking and building

Plan with AI.
Build with clarity.

You shouldn't have to re-explain your project every single time.

Open Pith

Free. Bring your own API key.

PITHv0.1

Add token counter to composer

Here's the implementation. Counts tokens client-side using tiktoken-lite.

typescriptcopy
// utils/tokens.ts
export function count(text) {
  return encode(text).length
}

via Sonnet 4.6

scratch.ts
1  import { encode } from 'tiktoken-lite'
2
3  export function count(text: string) {
4    return encode(text).length
5  }
6
7  // ~12k tokens (6%)
scratch.tstypescript
Sonnet 4.6Plan a feature, ask a question…
What makes it different

Claude has no memory. Cursor has no strategy.
Pith has both.

The three things that change how you build.

Prompt enhancer

Type rough, hit Cmd+L. Pith rewrites your prompt for maximum AI clarity before sending.

Project memory

Your stack, decisions, conventions — saved once, injected into every single message automatically.

Multi-model, your keys

Claude, GPT-4o, Gemini, Groq, local Ollama. One interface, your own API keys, zero markup.

The flow

Three steps. Every session.

Step 01

Set your context

Describe your stack once. Pith injects it forever.

Step 02

Enhance your prompt

Rough idea in, precise prompt out. One shortcut.

Step 03

Plan, then execute

Copy Claude's answer to Cursor. Ship.

What's included

Everything. No subscription.

Named sessions with history

Your planning conversations, organized and searchable. Never lose a decision again.

Sessions+ New
  • Token counter design2d
  • Onboarding flow2d
  • Voice input fix2d
  • Pricing page copy2d

Always-on rules

Injected into every response in this project.

Always use TypeScriptNo em dashesTailwind v4RSC by default

Local Ollama support

Desktop app with free inference. No API costs. Your machine, your models.

llama3.1:8b · running locally

Session export

Save any conversation as markdown.

Token counter

Know your context usage.

~12k tokens6%

Voice input

Speak your prompt. Transcribed by Whisper.

Multi-project

Separate context, rules, and history per project.

Pith
  • overnight.so
  • New idea

Built for where this is going.

AI tools are about to get much more powerful. Pith is being built to grow with that.

Coming soon

GitHub integration

Pull your actual repo context automatically. No more pasting architecture docs.

Coming soon

Team memory

Shared project brain for engineering teams. One update, everyone's AI gets smarter.

In research

Decision capture

When Claude recommends something and you approve it, Pith saves it as a logged decision. Searchable. Referenced forever.

Everyone has a plan.

Nobody saves it.

Pith remembers everything

so you don't have to.

Because context rot kills momentum.

FAQ

Questions, answered.

It goes to your own server (Vercel) to make the API call, then to Anthropic/Google/OpenAI directly. Never stored, never logged.

Pith saves your project context, rules, and session history permanently. Claude.ai starts fresh every time.

Yes with Ollama. Pull a local model and plan without any internet connection or API costs.

Yes. Pith supports Claude, GPT-4o, Gemini 2.5, Groq Llama, Grok, and local Ollama. Swap mid-conversation.

Coming soon. Shared project context for engineering teams. Join the waitlist.

Stop explaining your project from scratch.

Set it once. Every conversation knows everything.

Open Pith

Free. Your own API key. No subscription.