Agents

More ways to run

Codebase map

No map yet for this project.

System prompt (optional)

User prompt

Models — runs in parallel

Open a project, then click + New thread to start a conversation. Each thread keeps its own context, token budget, and cache prefix — switch between tasks without polluting any single chat.

Ask anything about this codebase

Index maintenance

TrustModel.ai Observability + Guardrails + Compliance

Generate API Key Open Control Plane

Trace

Mission Control

Dispatch tasks to agents and supervise them as they run — concurrently, asynchronously, in one board.

Build Agents

Describe a recurring task in plain English. AgentFoundry drafts a plan you can review, runs it on your machine against your own logins and files, then saves the result as a reusable agent. Nothing leaves your machine without your approval.

Examples →

Advanced ▾

Working folder (optional, local files only):

Loading model selection…

My assistants

No assistants saved yet. Describe a task above and run it once — you'll be able to save it from there.

Your Agents

Skillify

Teach an imported agent how your company does things. Skills become part of the system prompt the agent receives on every test invocation.

Pick an agent

Teach it something new

When the agent encounters:

It should:

It should NOT:

Example from your experience (optional):

Test it

Sends a real request to the agent's endpoint, with the system prompt assembled from skills + guardrails.

Project knowledge (Wiki)

Drag markdown docs or folders here. Each .md file becomes a wiki page that every coding agent (Claude Code, Codex, Gemini, SMART-Code) sees at session start. No upload, no embedding pipeline — markdown files agents navigate by link.

Drop .md files or folders here, or

Test the knowledge (Retriever)

Ask a question; see which wiki pages + repo-map symbols the agent's retriever returns. Sub-100ms; zero embedding cost.

Guardrails

Configure what an agent may and may not do. Saved guardrails are sent to the agent as a system prompt on every test invocation. Runtime enforcement requires your agent's host to honor it (e.g., Agent21 / TrustModel.ai).

Pick an agent

DefenseClaw runtime checking…

Cisco's open-source agent-security framework (Apache 2.0). Wraps the agent's runtime with Skills Scanner, MCP Scanner, CodeGuard, and a Guardrail Proxy. AgentFoundry shells out to your local defenseclaw CLI — nothing is mocked. Repo →

Agent Fleet

Every agent you've built — across every project — in one pane. Run, inspect, fix without losing your place.

Sidecar deploys

For agents running in your own cloud (AWS/GCP/Azure) under strict security. Install a small daemon inside the VPC; it long-polls codenow.pro and serves the latest skills + guardrails to your agent over loopback. How it works ↗

Open a shared project to mint a sidecar token.

Output

Real test invocations from this AgentFoundry instance. Each row shows the actual prompt sent and response received.

AgentFoundry

The Vibe-Coding IDE

Add your first project

Create a brand-new project, or open an existing folder. Each project becomes a tab.

You can add more projects anytime with the + button

AgentFoundry

Connect at least one AI provider

Add your first project