Arbiter blog
The Arbiter engineering journal
Engineering notes from the working orchestration brain: durable recovery, model routing, scheduling around throttling, and what it takes to dispatch an entire AI operation from one agnostic command center.
One brain on your card, shared by the whole stack
Every local-first tool wants its own model on your GPU, and a single card runs out of room to hold them all. Arbiter keeps one resident brain and lets the rest of the stack borrow it, so Curator's memory work and Arbiter's own routing run on a single loaded model instead of a pile of duplicates.
Read post →Take the wheel of your own agent sessions, from anywhere, behind a second factor
Arbiter can drive your own live coding sessions remotely, including running work on them. Reading stays inside your own private bot channel, and every acting step is gated further by a single-use code from your authenticator. Here is how that stays safe.
Read post →Arbiter's local brain just got more room to think
The on-device model that runs Arbiter's orchestration was upgraded. It now holds more of your operation at once and returns results in a shape you can rely on, all without a cloud call in the decision path.
Read post →Supervise a fleet of coding agents, not a single window
One agent in one window is easy. A fleet of them is where the real work is, and where the tools stop helping. Arbiter launches, names, and drives many live sessions at once, with status you can trust about which ones are actually alive.
Read post →Route every task to the cheapest model that can actually do it
Frontier models get handed trivial work every day, and the bill never notices. Arbiter routes each task against the capability, cost, and safety policy you set, and absorbs the coordination work on a local brain that costs nothing per call.
Read post →Own the terminal. Don't live inside it.
Most AI coding work happens inside one terminal window at a time. Arbiter treats the terminal as a system layer it owns and drives from above, which is what lets one operator run a fleet instead of babysitting tabs.
Read post →Cheap by Default, Controlled From Your Phone
Make capability, cost, and safety tradeoffs explicit in routing policy, then use configured remote approval for sensitive work. Arbiter is a local-first working alpha in public binary release preparation.
Read post →Approve Your Agent's Plan From Your Phone, Then Let It Run
With a supported remote channel and approval policy configured, review an agent plan from your phone, approve sensitive execution with a single-use code, and follow the session.
Read post →Guardrails for Autonomous Agents
Fanning an agent out across many workers is powerful and risky. Arbiter provides configurable controls for stopping work, limiting depth, setting budgets, and requiring approval, while interrupted side-effecting work remains visible for review.
Read post →What Is Remote Agent Supervision?
As developers hand more work to long-running AI agents, a new problem appears: supervising agents you cannot watch, from wherever you are, without shipping your keys or your code to someone else's cloud. Here is what good looks like.
Read post →Headless follow-ups into Claude and Codex
With a supported remote channel configured, send a running Claude or Codex session more work without opening a window or losing its context. Arbiter reaches the session it already has.
Read post →Remote control, locked down
Arbiter lets you drive your AI operation from your phone. That power needs a lock on it, so remote control sits behind a per-conversation code from your own authenticator.
Read post →Your standup, run by a local brain
Arbiter's local brain can support the operations loop: a morning briefing across projects, a tool-calling chat, and configured dispatch to coding agents with policy-controlled approvals and stopping controls.
Read post →Arbiter is alive: the orchestration layer, now running
Arbiter is a working proprietary alpha: a local brain on your own GPU, configured multi-provider routing, durable interruption handling, and configurable remote control. Public binary release preparation is underway, and there is an 8-minute walkthrough.
Read post →Arbiter's brain runs on your GPU
Provider-agnostic should not stop at the providers. Arbiter can use a local model on your own hardware for routing assistance, without requiring a cloud call in that decision path.
Read post →You already have an AI fleet. You just don't have a commander.
Arbiter is a local-first orchestration layer for configured agents, providers, and jobs. It can apply routing policy, preserve durable run state, and use a local GPU model for orchestration assistance.
Read post →Provider-agnostic orchestration: why we don't lock you in
Why an orchestration layer should stay neutral about the models, clients, and accounts you bring to it.
Read post →Introducing Arbiter: the orchestration layer your AI stack is missing
A provider-agnostic command center for configured LLM clients, terminals, and long-running jobs. It applies routing policy and preserves interrupted run state for deliberate recovery.
Read post →