Get Started in 5 minutes

Pick your platform

Install

Open the DMG, drag CodeNow.app to Applications, then launch from Applications or Spotlight (Cmd+Space → CodeNow). The app is signed by Developer ID Application: PREDIXTIONS INC and notarized by Apple — first launch should be Gatekeeper-clean (no “unidentified developer” warning).

Slack

Click Slack → OAuth to your workspace. Connection is per-project; the AI in this project gets read access to the channels you authorize.

GitHub

Click GitHub → install the GitHub MCP. Used for issue context, PR diffs, repo-aware AI suggestions.

Jira

Click Jira → connect via Atlassian OAuth. Tickets become first-class context for AI prompts.

Custom MCP

Anything that speaks the Model Context Protocol can be wired up via + Add Custom MCP Server — Notion, Linear, custom internal tools.

Open the bottom terminal

Press Ctrl+` (backtick). The terminal panel appears with a real shell — full system access, no sandbox.

Start the AI assistant

Click Claude Code tab in the bottom panel (or run claude directly). The assistant runs against your active project’s files. Drag any file from Finder onto the terminal to paste its absolute path — saves typing.

Ask

“Create a small Express app that serves a hello world page on port 3000 and adds a /health endpoint that returns the current time. Then run it.”

The AI writes the files, hands you the run command. When you’re done, deploy with one of the built-in deploy commands — vercel for Vercel, gh for GitHub releases, or any of the standard CLIs.

Switch models if you want

Click the model switcher at the top — Claude Code, OpenAI Codex, Gemini, Aider, Copilot. Per-project, so each project remembers its preferred model.

Open the Agents sidebar

The Agents tab in the sidebar shows the agent lifecycle: Anatomy · Run · Test · Runs · Trace · Evals · Tools · Prompts · Skills · Publish.

Describe → Build → Run

Type what you want in the “What should your agent do?” box and click Build. That’s it — CodeNow drafts the agent, scaffolds it (auto-picking a “Recommended for you” kit from the keys it detects), and lands you on a running agent. No SDK choice, no orchestrator/harness/MCP jargon up front.

Want to build step by step?

Prefer full control? Open the “Advanced: build step by step” disclosure to launch the 9-step guided builder — kit choice, sub-agents, deploy targets, and the harness/orchestrator/vendor-lock-in explainers all live there, behind disclosures. It’s opt-in; the Describe → Build path is the default.

Run it

Type a message in the Anatomy tab’s chat box and hit Enter. CodeNow runs the agent and captures the run as a recordable timeline. The response appears as a chat turn with cost + duration.

If you scaffolded an SDK kit and want to run its code directly from the terminal, drop the one env var its START_HERE.md names into .env.local (already gitignored) and run npm run agent:dev (TS) / python agent.py (Python).

Replay it

Open the Runs tab. Every run is there with cost, latency, status. Click any run to step through the timeline — every tool call, every model response, every token. Click Fork to current code → to re-run the same input through your current local agent and diff side-by-side.

Add a tool

Open the Tools tab → + New Tool. The Visual Tool Builder takes a name + schema and emits SDK-correct boilerplate (tool() + createSdkMcpServer()). Drag the new tool from the Tools tab onto the Anatomy card → it’s added to manifest.tools.

Publish

Once it works locally, Publish → Connect Agent21 → Publish packages your manifest (prompts + evals + tool signatures + guardrails) and POSTs to the Agent21 marketplace under your team’s branding.