Self-hosted AI agents: Elixir/OTP enters the race for lightweight, sovereign infrastructure

By Hacker News · July 2, 2026. Auto Learning Agents is billed as a self-hosted AI agent platform, packaged into a single Docker image that includes the Elixir runtime, Python services, a local database and a tooling layer. Installation boils down to four steps: install Docker, clone the repository (hosted under AIAppsAPI/auto-learning-agents), add API keys for providers such as Anthropic, OpenAI or Google —or skip them entirely and run local models via Ollama— and spin everything up with `docker compose up`. The project is advertised as free, open source and designed so the user 'owns the entire stack'.

The most notable technical detail is the choice of Elixir/OTP as its foundation. OTP (Open Telecom Platform) is the supervision framework of Erlang's virtual machine, originally designed for telecom systems that cannot afford to go down: lightweight processes, isolated from one another, with supervision trees that automatically restart whatever fails. Applied to autonomous agents, the conceptual fit makes sense: an agent that runs tasks independently and can fail mid-task benefits from a 'let it crash' model with automatic restart, rather than the manual exception handling typical of pure Python-based stacks. It is no coincidence that the project itself frames this as fault tolerance among its documented capabilities.

That said, we should be honest about what we're looking at: a documentation and installation page, with no adoption metrics, no real-world use cases described beyond generic mentions (customer support, scraping, content generation), and posted on Hacker News under the 'Show HN' format with a single point and zero comments. That doesn't invalidate the project —many serious launches start this way— but it does call for caution: there is still no public evidence of performance, audited security, or comparisons against established frameworks like LangGraph, CrewAI or AutoGen, which the project's own site lists as references in its documentation.

Overall, the agent-infrastructure sector is going through a moment of proliferation: every week new frameworks appear promising orchestration, persistent memory and tool execution, almost all of them with the same pitch of 'self-host your own AI and control your data'. That fragmentation is, at bottom, a healthy sign: it reflects that building reliable agents still has no dominant standard, and that the field is open to both giants and small projects with a solid architectural idea. The choice of Elixir/OTP fits well with the technological sovereignty thesis we've been defending —running your own stack, with your own keys or local models via Ollama, without depending on a closed provider— but real validation will come from adoption, benchmarks and production use cases, not from a well-written installation page. For now, this is a project to watch, not one to recommend without further evidence.