Changelog

Manage your machines from a terminal — or from an agent. npm i -g @b3dotfun/b3iq-cli, then b3iq login signs you in through the browser (approve a short code in the dashboard; no password ever touches the terminal). From there: b3iq status, machines list, models install <model> --machine <m>, runtimes restart, logs --follow, and b3iq ssh <machine> with the transport handled for you. Output is agent-friendly by design: stdout is data, stderr is logs, JSON is automatic when piped, and exit codes are stable. Tokens are personal access tokens you can see and revoke under Settings → CLI tokens. See b3iq CLI.

Every machine page now has an SSH access card: add public keys per machine or once for your whole account (Settings → SSH keys — installed on every machine you own and revoked everywhere when removed), open a browser terminal straight from the dashboard, or enable the native SSH lane and use plain ssh/scp/ rsync from your own terminal. Key changes reconcile to machines automatically.

The dashboard Models page now aggregates every model across your machines with real usage, installs a model onto several machines at once, and lists only models that are actually healthy to serve. Under the hood the host agent got stricter: model installs and removals restart the serving engine and verify the catalog actually changed, restarts report honestly instead of silently doing nothing, and a watchdog now revives a managed engine that crashes — the machine page shows a "won't self-heal" badge on the rare engine outside watchdog coverage.

You now get an email when one of your machines goes offline, and optionally when a GPU runs hot or a disk fills up (Settings → Machine alerts; on by default, with an override address). The dashboard also gained a Support tab — open a ticket, attach files, and track replies from the team without leaving the app.

You can now configure and purchase a dedicated GPU machine directly on b3iq.org — pick a platform, complete checkout by card, or book a call with the team; financing options are available at checkout. Enrolling a machine you already own now uses an onboarding code from the b3iq team, and pre-onboarded machines are linked to your account by redeeming a claim code in the dashboard. Machines can also be owned by an organization, with member roles deciding who can manage models, keys, and SSH.

Verified host-agent updates now roll out automatically across your fleet. Every machine keeps the signed-update + health-check + automatic-rollback pipeline, and each machine page has an auto-update toggle plus a manual apply path if you prefer to hold updates on a box. See Updates.

The dashboard playground now builds its model list from the machines on your account. Models are auto-discovered from what your machines actually serve right now, and you can pin a run to a specific machine or let routing spread across all of yours. The API matches: GET /v1/api/models called with your gateway key returns only the models your machines serve (any node scope on the key is respected), while anonymous calls keep the public network-wide view. The dashboard playground and your key's model list no longer show a global catalog — they show your machines. See Hosted Gateway.

Hosted inference no longer bills anything. Your API keys route requests to the machines on your own account, free of charge — no credits, no prepaid balance, no top-ups. Usage is still metered per key and account, and accounts carry an optional daily token quota (default uncapped). When a request can't be served by any machine on your account it fails closed with no_eligible_nodes. Route-ready machines earn USD from B3IQ-directed workloads, with payouts unchanged. See Usage & Quotas and Claiming and earnings.

One hosted chat request can now fan a prompt out to several models and return one combined OpenAI-compatible answer. Attach a b3iq.mixture object choosing a strategy — synthesize (an aggregator model writes the final answer), vote (self-consistency plurality), or weighted (best-of failover) — with per-expert weights, a quorum, and an aggregate token budget checked before any node runs. Works free across the machines on your account, including with a fleet key. See Mixture of Experts.

Operators can keep an enrolled node out of the public pool and run it in an owner-private network instead. A per-node and bulk toggle switches a node between public routing and your private fleet, so you can serve traffic across nodes you own without joining the shared pool. (Superseded in July 2026: the Public/Private toggle is gone — own-nodes-strict routing is now the default for every account, and the per-machine choice is Bare metal vs Earn mode — see Run a private fleet.)

Linking a node to your account now uses a short on-screen claim code instead of the older multi-step enrollment flow. Scan the QR or type the code into the dashboard to pair the node, then continue setup. The host agent still owns its own local identity and signing; the code only binds an already-running agent to your account.

The setup wizard can install verified inference runtimes for you instead of asking you to provision them by hand. On Apple Silicon Macs the default is MLX, with a llama.cpp Metal fallback when MLX is not a fit for the selected model. You can still install and point at your own runtime if you prefer.

Node software updates are signed and can be applied with one click from the dashboard. An update runs a post-apply health check and automatically rolls back if the node does not come back healthy, so a bad update should not leave a node stuck offline.

Setup and day-to-day operation now live in one place. The guided setup walks a new node through pairing and readiness, and you can pause and resume serving from Settings without tearing down the node. The split between a separate setup flow and an operations view is gone.

Developers can top up prepaid credits by card (Stripe) or crypto (AnySpend), and the credits and usage views are clearer about current balance and recent spend. Billing still fails closed when balance is insufficient. (Superseded in July 2026: inference is now free on your machines and credits no longer exist — see Usage & Quotas.)

A desktop companion tray app is in pilot on macOS (signed, notarized DMG) and Windows (Authenticode-signed NSIS installer), both downloadable from b3iq.org. It is an early, limited rollout, not a general release, and the host agent and dashboard remain the supported path for running a node. Linux stays on the install scripts.