Aleph
3
Sign in
[ Index — 00 ]Local-first · One core, many shells · Multi-agent
Est. 2026 — Aleph LabsN 37.77° · W 122.41°
Universal Personal AI

Aleph

Your agent intelligence — not a black box.

One personal agent that lives natively on your desktop and reaches you on every channel you already use — Telegram, Slack, your browser. It remembers you, runs on your terms, and quietly orchestrates a team of specialists across research, writing, code, scheduling, and the small logistics of a life.

Also for macOS · Windows · Linux
01 Runs locally by default02 Remembers across every channel03 Orchestrates sub-agents
Scroll to explore ↓
[ Philosophy — 01 ]Aleph / ℵ — the first letter

We treat your assistant as a partner you can read, not a box you have to trust.

Most AI hides its reasoning behind a cloud URL and hopes you don't ask. Aleph does the opposite. The brain runs on your machine; every channel — desktop, phone, browser — is just a door to the same mind. Every plan it forms, every sub-agent it dispatches, every memory it keeps, every tool it runs in its sandbox is legible, editable, and yours. Intelligence that earns trust by showing its work.

Legible see every stepLocal one core, many shellsSovereign your keys, your sandbox
Fig. 1 — the legible mind

Capabilities — 02

Skills, systems & an interface — all in service of one agent.

Pick a capability to preview how Aleph actually handles it — live, step by step. Ten specialists, one conversation, zero context-switching. Every step is visible; nothing runs off your device without your say-so.

aleph · deep research
Compare the three leading approaches to local-first agent memory. Cite sources.
Aleph · working
Searched 11 sources
Cross-checked 3 with citations
Drafting comparison table…
AlephThree approaches: vector recall, event-sourced logs, and hybrid graphs. Hybrid wins on traceability — full reasoning + 6 citations attached. Open the trace?

Skills archive — 03

A living archive of skills your agent keeps learning.

Browse all skills

How it works — 04

From a signal to a finished system.

You give Aleph an intention. It handles the rest as a transparent, four-beat loop you can pause and steer at any point.

01

Capture

You hand over an intention — a sentence, a file, a voice note.

02

Understand

Aleph reads context from memory and forms a legible plan.

03

Orchestrate

Sub-agents run in parallel in a sandbox; you can pause, steer, or veto any one.

04

Deliver

A finished artifact, plus the full trace of how it got there.

Agents — 05

Turn a vague brief into a memorable, deliverable artifact.

Aleph spins up purpose-built sub-agents on demand. Each one is a named, inspectable specialist you can reuse, share, or retire.

12+ built-in agents custom skills
aleph/guzangdecks
live

Builds argued, on-brand presentations from a one-line brief.

aleph/kamidocs
live

Captures messy thinking and returns a structured, sourced doc.

aleph/atlasresearch
live

Runs deep, multi-source research with a traceable reasoning log.

aleph/conciergelogistics
live

Handles travel, orders, and errands end to end — you just confirm.

[ Field notes — 06 ]From early access

“Aleph turned my fuzzy AI wishlist into a system that’s sharp, trustworthy, and genuinely mine.”

Mira Adelstein
Independent researcher
HelixNorthbeamKernáBYVÖKAtlas&Co
Fig. 6 — partner, not tool
Engine room — 07

Powered by the world’s best agents & models — pick yours.

Aleph is model-agnostic. Route any task to the engine you trust — or let Aleph route each step dynamically to the right model. Swap providers anytime without losing your memory or your skills.

ClaudeGPTGeminiLlama (local)MistralQwenDeepSeekYour own

Questions — 08

On Aleph, agents, memory & the local-first question.

Talk to us

By default, Aleph runs on your device with your data. Memory, files, and reasoning stay local. Risky tools run in an OS sandbox and ask before acting — and when you explicitly route a task to a cloud model, you see exactly what leaves your machine.

A chatbot answers in one window. Aleph is one brain that lives natively on your desktop and reaches you everywhere — Telegram, Slack, your browser — with the same memory and identity. It dispatches named sub-agents for research, writing, code, and logistics, runs them in parallel, and stitches the results into one finished artifact — all in a single conversation.

Every plan, every sub-agent call, and every memory Aleph keeps about you is visible and editable. You can open the trace on any task, see the reasoning, and correct it. Nothing happens in a black box.

Yes. Aleph is model-agnostic. Route any task to Claude, GPT, Gemini, a local Llama, or your own fine-tune — and switch providers anytime without losing your memory or your skills.

Aleph keeps a private, structured memory — linked notes it can traverse, not just a vector blob — of your preferences, projects, and people, compacting and connecting them over time. It is yours: inspectable, exportable, and deletable at the level of a single fact.