AI toolkit, not a platform

Equip your AI agent for code review.

A harness bundles your model, MCP servers, skills, and a live sandbox. Switch contexts in one click — your agent gets a different brain for every job.

Get Started See it in action

Bring your own MCPsDaily + weekly cost capsLive sandbox built in

Issue triage agent·claude-sonnet-4.6

3 MCPs

Pull this week's open auth-related bug reports and file the highest-signal ones in Linear.

github·search_issues("auth is:open")142ms

linear·create_issue("Auth: SSO token refresh")268ms

Found 7 open issues mentioning auth. The top theme — SSO token refresh failing — appears in 4 reports. Filed LIN-481 and grouped the long tail under LIN-482.

/linear_create_issue↵ run

Features

Built for the agent you actually use. Not the demo your boss saw on Twitter.

One agent, many contexts

A harness saves your model, system prompt, MCPs, skills, and sandbox. Switch from a coding agent to a research agent in one click.

Plug in any tool

A growing catalog — GitHub, Notion, Linear, Exa, Context7, and more. OAuth handled. Custom MCP URLs welcome.

Real code, not just chat

Each harness can attach a Daytona-backed sandbox: file explorer, terminal, git. Your agent edits, runs, and ships — you watch it land.

Skills from skills.sh

Pull battle-tested instructions for code review, debugging, web search, PDFs. Your agent learns your team's playbook on import.

Slash commands & rich attachments

Every MCP tool becomes a / command. Drop in images, PDFs, audio. Streaming reasoning across Claude, GPT, and Gemini.

Transparent budgets

Daily and weekly cost caps with per-model and per-harness breakdowns. No surprise bills at the end of the month.

The Product

Three real surfaces. No mockups.

This is what you'll see five minutes after signup — the same UI we use every day.

MCP Status & Add

Add servers without leaving the chat.

Click the MCP indicator in the header. Pick from the catalog or paste a custom URL. OAuth pops automatically. Status badges update live.

MCP Servers

GitHub

Connected

OAuth

Linear

Connected

OAuth

Context7

Connected

Public

Notion

Token expired

Slash Command Palette

Every MCP tool. One keystroke away.

Type / to filter across every connected server's tools. Keyboard-first, typo-forgiving, scoped to the active harness.

/githubcommands

github_search_repositories

Search public + your repos

↵

notion_create_page

Create a page in a database

linear_create_issue

Open an issue in your team

context7_fetch_docs

Pull up-to-date library docs

exa_search

Semantic web search

Live Sandbox

A real dev environment, attached to the chat.

Daytona-backed sandboxes give your agent file editing, a real terminal, and git — persistent across sessions. Watch tests pass before the agent commits.

Files

Terminal

Git

sandbox · running

$npm test

→Test Suites: 12 passed, 12 total

→Tests: 148 passed, 148 total

✓All green. Committing.

$git commit -m 'feat: streaming retries'

→[main 2f1e9a0] feat: streaming retries

How It Works

Three minutes from signup to first answer. No SDK to install.

Configure

Name a harness. Pick a model — Claude, GPT, or Gemini. Attach the MCPs and skills your agent needs. Optionally add a Daytona sandbox.

Connect

OAuth into GitHub, Notion, Linear, or any provider in one popup. Tokens are stored encrypted and refreshed on demand — no manual key wrangling.

Chat

Streaming reasoning, slash commands, tool calls, attachments, and live sandbox output — all in one conversation.

Build your first harness.

Free to try. Bring your own MCPs and your own model preferences. Three minutes to your first answer.

Get Started See features