AgentFoundry Wiki

The user guide, get-started flow, and agent concepts for AgentFoundry — the dual-mode IDE for shipping apps and AI agents.

v0.1.68 is live. The full IDE is now free for everyone — no trial, no expiry. Build an agent by describing it in plain language and it scaffolds real code and runs keyless. Read the release notes →

15 things AgentFoundry does that Cursor, Windsurf, and Codex don’t:

Multi-project tabs in one window — open every codebase you work in side-by-side. Each tab keeps its own AI model, MCP connections, terminal, agent state. The others are all one-project-per-window.
Dual-mode IDE — Developer mode for engineers, Business mode for PMs / ops / sales / legal. Same window, same project state.
Multi-model side-by-side — Claude, Codex, Gemini, Aider, Copilot in one window, BYOK, switch in one click.
All four SDK scaffolds — Claude Agent SDK (TS + Python), Cursor SDK, Microsoft Agent Framework, Google ADK. Build for any provider.
AgentFoundry Sidecar (Mode C, v0.1.28) — push skills + guardrails into agents deployed in your own AWS / GCP / Azure cloud. Outbound HTTPS only; no inbound firewall holes. See Deployment modes →.
Live Agent Replay — every agent run captured as JSONL, replayable, forkable to current code with diff.
Skillify + Guardrails — business users train any agent from a no-code UI, enforced as system prompt on every invocation.
CRDT pair-programming (v0.1.28) — real cursors, real-time edits in the same buffer. No screen-share. See Pair programming →.
JIRA → multi-model context — click a ticket, full context (summary, priority, labels, description) injects into every selected agent.
Embedded browser panel — AI navigates, screenshots, clicks, evaluates JS in a real WebContentsView. Locally, no cloud computer-use service.
One-click deploy — Vercel / AWS / Azure / GCP / Kubernetes from inside the IDE, with a signed TrustScore artifact + GitHub Action wrapper.
Per-project MCP connections — 49 vetted integrations (GitHub, JIRA, Slack, HubSpot, Salesforce, Postgres, Stripe…) scoped per project. The others only support global MCP.
Anonymous read-only share link — mint a URL, recipient clones and opens the project in viewer mode. No sign-in required. Revoke any time.
Mobile companion at codenow.pro/app — review runs, approve trust gates, edit skills from your phone. PWA, no app store.
Agent21 marketplace publish — one-click bundle (prompts + evals + tools + guardrails) and list your agent for sale. Built-in monetization.

Wiki contributions welcome at github.com/karlmehta/codenow-wiki . The IDE source lives at github.com/karlmehta/codenow ; the wiki is open to community PRs without exposing the IDE internals.

Start here

The coding assistant inside AgentFoundry — two AI brains, real file reads, no black box. Start here if you’re new.

SMART Code Assistant

Download → install → connect Slack/GitHub/Jira → ship your first app and your first agent.

Get Started in 5 minutes

What is an agent? Skills, tools, connectors, subagents, runs, replay — explained with diagrams.

Agent concepts

The vibe-coding flow: project tabs, multi-model AI, real terminal, real deploys.

Build your first app

Describe → Build → Run — no API key needed. Or take the Advanced path to pick a harness (Claude Agent SDK, Cursor SDK, Microsoft Agent Framework, Google ADK).

Build your first agent

Region-by-region tour of the IDE. Every clickable element documented with screenshot, shortcut, and what it does.

UI map — every button

Why AgentFoundry

AgentFoundry is what happens when an IDE is built for the agent era from the ground up — not a code editor with AI bolted on.

Multi-model, multi-harness — Claude Agent SDK · Cursor SDK · Microsoft Agent Framework (Azure) · Google ADK (Vertex). Bring any harness; we ship the lifecycle around it.
Live Agent Replay — every run captured, replayable, forkable. Production traces in the same surface as dev.
Trust + Safety built in — observability, guardrails, evals, HITL approvals, audit trail. No SDK install .
Dual-mode — engineers ship the agent, domain experts skill and guardrail it, operators monitor it. All in one IDE.
Publish to Agent21 — one click, OAuth-authenticated, full provenance manifest.

Wiki sections

SMART Code Assistant — the coding assistant inside AgentFoundry
Get Started — install, configure, first app, first agent
Concepts — agents, harnesses, runs, replay, manifests
Build — guided builds for apps and agents
Connect — Slack, GitHub, Jira, custom MCP
Reference — UI map of every button, keyboard shortcuts, file layout
Troubleshooting — common install + runtime issues

External resources

This wiki adapts and references several open-source bodies of work:

Agentic Design Patterns by Evoiz — open-source book on agent patterns. We translate the patterns through the AgentFoundry vocabulary (Anatomy card, harnesses, manifest model).
Claude Agent SDK docs — Anthropic’s official docs.
Anthropic’s “2026 Agent” mockup — the visual mental model behind AgentFoundry’s Anatomy card.

The wiki is released under the same open license as the Agentic Design Patterns source where adapted; AgentFoundry-specific content is MIT.