Agentlas
Describe a worker, get an AI agent that does it
3 followers
Describe a worker, get an AI agent that does it
3 followers
Every agent SaaS wants your API keys, your chat history, and a second monthly bill. Agentlas refuses by architecture. ZERO models run here; nothing is ever proxied. Every call hits Claude, OpenAI, Google, or Ollama straight from your machine, on the plan you already pay for. Keys in your keychain, agents in local SQLite, whole teams as live org charts with memory baked in PM Soul, Memory Curator, Task Bias. GUI or terminal, one runtime. Open source. Your machine, your rules.
4 months later Agentlas has grown into something much bigger than the agent builder we originally launched here.
The original idea was simple: describe the AI worker you need, and Agentlas builds it.
Today, we're building the local operating layer for AI agents.
Agentlas Desktop lets you run Claude Code, Codex, Gemini, API models, and local models from one place ā using the subscriptions and API keys you already have. Agentlas never proxies the model call. Your keys stay in your OS keychain, your chats and agents stay local, and the core runtime is open source.
But the bigger change is that Agentlas is no longer just about creating a single agent.
You can build or borrow specialists, organize them into teams, route work between them, give them tools and memory, run AI-native Apps, automate recurring work, and inspect the team and files behind every run.
Agentlas OS is the open-source engine underneath it. It works across supported AI runtimes rather than locking an agent to one model vendor.
The direction we're pursuing is simple:
Build an agent once. Own it. Run it with the models and computers you choose. Borrow specialists when you need them. Combine them into teams when the work gets bigger.
We think AI agents should be portable assets not configurations trapped inside one model provider.
Agentlas Desktop is free and open source, and Agentlas OS is open source under Apache-2.0.
A lot has changed since our first Product Hunt launch, and we're still shipping fast. Feedback, criticism, and GitHub issues are very welcome.