Show IH: I built an open-source AI agent manager — and an AI agent is running the marketing

Hey IH 👋

I've been lurking here for a while and finally have something worth sharing. I built Jam — an open-source desktop app for managing multiple AI coding agents from a single interface.

The itch I was scratching

I use AI coding agents daily — Claude Code, Codex CLI, Cursor. The problem: I'd end up with 4-5 terminal windows open, each with a different agent, none of them sharing context. Every new session started from scratch. I was copy-pasting project background between tabs like it was 2015.

I wanted one place to:

• Run multiple agents simultaneously (each in their own terminal)
• Talk to them by name using voice ("John, refactor the auth middleware")
• Have them remember past conversations so I don't re-explain everything daily
• Watch all of them work at once in a grid view

Nothing like this existed, so I built it.

What Jam does

Jam is mission control for AI agents. You create agents, assign them a runtime (Claude Code, Codex CLI, OpenCode, or Cursor), point them at a project directory, and let them work in parallel.

Key features:

• 🤖 Multi-agent orchestration — Real PTY terminals, not API wrappers. Full tool use and file editing
• 🎙️ Voice control — Name-based routing with Whisper STT + ElevenLabs/OpenAI TTS. Each agent gets a unique voice
• 🧠 Persistent memory — Conversations saved as daily logs. Agents pick up where you left off
• 📝 Living personalities (SOUL.md) — Each agent has an evolving personality file that learns your preferences
• 🔧 Dynamic skills — Agents auto-generate reusable skill files from patterns they discover
• 🖥️ Stage view — Grid showing all agents working simultaneously

Tech: TypeScript monorepo, Electron + React + Zustand. Cross-platform (macOS signed & notarized, Windows, Linux). No cloud, no accounts, no telemetry.

Demo: https://youtu.be/sXrvp5j5U6s

The meta part 🤖

Here's the thing that makes this real: one of my agents (named "John") is running the entire marketing campaign for Jam. John built a Kanban tracking dashboard, drafted content for Dev.to, Reddit, and Twitter, posted to Hacker News, and is managing everything end-to-end. Every task assigned to @john on the campaign board was executed autonomously.

That's the kind of workflow Jam enables — agents that don't just write code, but operate across your entire workflow with persistent context and real tool access.

Where things stand

• GitHub stars: 10 (very early days — launched last week)
• Revenue: $0 — this is a free, open-source tool (MIT license)
• Platforms: macOS, Windows, Linux
• Team: Solo dev + a team of AI agents doing the heavy lifting
• Built in: ~3 months of nights and weekends

My ask

I'd love feedback from this community on a few things:

1. Multi-agent vs single-agent — Do you actually use multiple AI coding tools, or is one enough?
2. Voice control — Gimmick or genuinely useful in your dev workflow?
3. What runtimes would you want supported next?

Try it

Pre-built binaries: macOS | Windows | Linux

Or build from source:

git clone https://github.com/dag7/jam.git
cd jam && ./scripts/setup.sh && yarn dev

GitHub: https://github.com/dag7/jam

If you find it useful, a ⭐ on GitHub means a lot at this stage. Issues and PRs very welcome.

Built by Gad — scratching my own itch of managing AI agent chaos. The irony of AI agents marketing an AI agent manager is not lost on me.