Dev blog

The Aleph blog

Here we write about how we're building it: decisions, mistakes, things that worked and things that didn't.

July 2026 · 7 min

How we test that Aleph is not fooling itself

We added a deterministic fake-model suite to test the real harness: tool calls, permissions, errors, loops, context, grammar and native.

Read
July 2026 · 6 min

Aleph learns to remember (without the cloud)

Persistent memory across sessions with project-scoped MEMORY.md and global USER.md.

Read
June 2026 · 7 min

Aleph learns to google (without Google)

web_search and web_fetch with zero paid APIs, plus detection of JS-only pages.

Read
May 2026 · 6 min

The agent that actually does things

Tools, permissions and the autonomous loop that turns the model into a real assistant, not just a chatbot.

Read
April 2026 · 5 min

Real streaming: token by token

Why streaming matters and how we wired llama-server to the UI in real time without anything blowing up.

Read
March 2026 · 6 min

10,000 models — which one do I download?

Topic chips, hardware badge and curated catalog from a JSON, with no recompile needed.

Read
February 2026 · 5 min

Why I built a local coding agent

The origin story: tired of the cloud, choosing a stack, and the first version that actually worked.

Read