One brain · many hosts

Buddi is thinking

What would you like to grow today?

Buddi is an AI that sprouts with you — one agent runtime behind a hosted web chat, a terminal collaborator, and a coding agent. Ask, and watch it take root.

Start growing See how it works

The architecture

One core, three builds.

Every product is composed, not forked, from a single buddi core. Build a capability once and every build inherits it.

Hover a lobe to meet each archetype — all three mount the same buddi core.

buddi core — agent loop, tools, approvals, skillsweb ↔ sandbox only — the CLI runs locallyamber — the earth-accent, used sparingly

Developed by IT Team

One brain → web, desktop & terminal

Which buddi is for you?

Same core, three archetypes. buddi is the hosted web face; co buddi is a desktop app on your machine; code buddi is the terminal coding agent. Both local builds work on your real files, behind an approval gate.

Browser · webThe Thinker

buddi

For: Everyone — brainstorming, writing, research, everyday chat.
Does: Chat & respond with artifacts and file upload; code runs in an isolated sandbox — never on your machine.
Strength: General knowledge work + safe in-browser code execution.

Cloud-hosted; no direct access to your local machine.

Desktop appThe Collaborator

co buddi

co buddi · ~/notes

⚠ allow write?
✓ once — wrote digest.md

For: Knowledge & ops workers who live near their files — docs, notes, data, tasks.
Does: A desktop app that reads/writes your local files and runs tasks in your folder, behind a once / always / deny prompt.
Strength: Local, private, single-user work with your real files. Your key, no metering.

A friendly desktop app with real local-file access — like a workspace on your machine.

Terminal · CLIThe Builder

code buddi

- return d
+ if not d: raise
✓ 3 passed

For: Software engineers.
Does: Reads/edits code across a repo, runs code host-direct, shells, worktrees & subagents, verifies tests.
Strength: Development velocity — refactors, debugging, multi-file work. Learns your rules via BUDDI.md.

Runs locally like co buddi — the difference is the job, in the terminal.

co buddi & code buddi both run locally · your machine, behind an approval gate

Feature	buddi	co buddi	code buddi
Primary interface	Browser / Web	Desktop app	Terminal (CLI)
Target user	Everyone	Knowledge / Ops	Software engineers
Primary action	Chat / respond	Files / docs / tasks	Edit files / run code
Core strength	Brainstorm & write	Local file & task work	Development velocity
Where code runs	Isolated sandbox server (no network)	Host-direct, on your machine (approval)	Host-direct, on your machine (approval)
Shared	— one buddi core —	Local · your machine	Local · your machine

Reach for buddi for general thinking, research, and chat in your browser.

Reach for co buddi for hands-on work with your own files, docs, and tasks in a desktop app.

Reach for code buddiwhen you’re building, refactoring, or debugging code.

Three builds, one loop

Same brain. Different hands.

You ask → Buddi thinks → it may use tools, run code, or ask you to confirm → it answers, and can leave a lasting artifact.

buddi · web

The hosted chat that runs the numbers.

Ask over your files and Buddi computes the real answer instead of guessing — code runs in an isolated sandbox, results stream back as prose and a versioned artifact.

Try it — type a question, or pick one below.

BuddiQ2-metrics.csv

What’s the Q2 total?

co buddi · desktop

A collaborator that tends your work.

A desktop app on your machine with real access to your local files. co buddi plans and does — but sensitive actions surface a first-class once / always / deny approval. The gate is the isolation.

Run the task, then approve or deny the write.

co buddi — Desktop · ~/notes

Workspace · ~/notes

tidy this week’s notes into a digest

Ready — run the task to begin.

code buddi · cli

The coding agent that keeps tests green.

Claude-Code-shaped: it reads and writes files, proposes a diff, then runs the code right on your machine to prove the change works before it moves on.

Give it a task, review the diff, then run it.

code buddi — repo

› add input validation to parse_date()
ready — give code buddi the task to see the proposed diff.

How code runs safely

Two paths. One promise.

The web sends code to a clean room and waits for results. The CLI runs it right here — behind an approval you control. Same safety, told two ways.

Web · buddino network

Code never runs in the app process. It’s shipped to an isolated runner with a hard wall-clock kill and no network — results come home by value.

Local · co buddi + code buddiapproval gate

No sandbox needed — code runs directly on your machine. The once / always / deny gate is what keeps you in control of every sensitive step.

Capabilities sprout from use

Grow new abilities.

Import a .skill, connect an MCP connector — and every build inherits it at once. Buddi branches as you do.

pdf-reader.skillgithub · MCPpostgres · MCPslides.skillfigma · MCPbrand-voice.skill

Plant one seed. Grow everywhere.

One brain, in the browser and in your terminal. Start a conversation and watch your work take root.

Start growing