What's New

2026-04-22 — API 4.6.0: Visual artifacts + silicon workers as first-class identities

The biggest server release since launch. Three new primitives plus a smoother Lifetime checkout, end-to-end validated from a fresh Windows test box before publish.

Visual artifact pipeline

• Workers can now produce images, video, audio, and PDFs as first-class outputs. Each artifact gets a stable URL at modelreins.com/s/<slug> — three-tier hosting (companion-local, modelreins-tunnel, or your own protowebb).
• The review queue renders artifacts inline so a human can preview the actual image before approving for publish. Purpose-built for generative-AI compliance workflows where someone needs to eyeball every output.
• Foundation for the upcoming creative-team use case: marketing departments generate brand assets at scale, your queue catches anything off-brand before it goes live.

Silicon workers as first-class identities

• New workers_registry table — every silicon worker has a persistent identity independent of heartbeat/presence. Declared capabilities, risk tiers (auto / audit / approve / session), audit trail, revocation.
• New /workers dashboard for register / list / revoke. New reserved platform tenant for internal first-party workers.
• Pairs with the Python SDK 0.2.0 below for the full developer experience.

Self-serve API key minting

• New /settings/api-keys flow lets users mint scoped keys themselves. Keys appear once via a one-time-view URL with 15-minute TTL. No support tickets, no raw tokens emailed, no tokens in logs.

Smoother Lifetime checkout

• The Lifetime $999 checkout flow is now a single clean hop to signup. (The generic /checkout/{plan} handler dispatches directly to the dedicated lifetime renderer when plan=lifetime.)

---

2026-04-22 — Python SDK 0.2.0: upload_artifact() + submit_review()

Two new methods, zero new dependencies, still stdlib-only.

Worker.upload_artifact()

result = worker.upload_artifact(
    data=png_bytes,
    content_type="image/png",
    filename="output.png",
)
# {'slug': 'abc123...', 'url': '/s/abc123...', 'size_bytes': 1234, ...}

POST a binary blob through the artifact tunnel, get back {slug, url, content_type, size_bytes}. Multipart-encoded manually so the SDK stays zero-deps.

Worker.submit_review()

worker.submit_review(
    type="generated_image",
    title="Q4 launch poster — draft 1",
    content="Auto-generated by image worker on 2026-04-22",
    preview="https://app.modelreins.com/s/<slug>",
)

Push content to the human-review queue with optional artifact preview. Used when a worker produces output that shouldn’t publish without a human eyeball (risk_tier=audit or approve).

Image generation example

examples/image_generation_worker.py ships the full pattern end-to-end: claim a job, call 1minai IMAGE_GENERATOR, upload result as artifact, submit to review queue. Drop-in starter for any text-to-image worker.

pip install --upgrade modelreins-worker to get 0.2.0. Requires API ≥ 4.6.0.

---

2026-04-22 — Node SDK 4.4.4: catch-up release

Three commits’ worth of fixes that hadn’t been published since 4.1.0:

• Playwright worker provider — wrap web UIs that have no API as first-class workers
• Server-side killswitch awareness — the Node worker poll loop now honors a server-side killswitch so the dispatch fleet can be stopped centrally
• Better rate-limit handling — more upstream rate-limit error phrases recognized + classified, so workers back off gracefully

npm install -g modelreins-worker@latest or use it via npx modelreins-worker@latest <name>. Artifact-pipeline parity with Python SDK 0.2.0 is coming in a separate v4.5.0+ release.

---

2026-04-18 — Companion 4.5.5: Windows first-run fixes

Three small but real wins for anyone installing on a fresh Windows machine.

• No more “stuck waiting for Ollama.” Earlier 4.5.3 set OLLAMA_HOST=0.0.0.0:11434 to enable cross-machine dispatch, but on Windows that bind ends up IPv6-only (Windows’ IPV6_V6ONLY defaults to 1), so the Companion’s own IPv4 probe timed out and the wizard hung. 4.5.4 reverted the default to 127.0.0.1:11434 — safe loopback on every platform. Cross-machine dispatch will come back as an opt-in tray toggle in a later release (it needs netsh firewall rules on Windows anyway, not just an env var).
• Double-click no longer spawns two installers. A double-click is the default Windows action to launch a shortcut. Without a mutex, the one-click NSIS installer spun up a second wizard on top of the first and the user saw what looked like two ModelReins installers running at once. 4.5.5 holds a named mutex on launch; the second click silently aborts.
• 256k context length is now the default for Ollama. No more silent 4k-default quality loss on long prompts. The Companion sets OLLAMA_CONTEXT_LENGTH=262144 before launching Ollama so the “context length” slider nobody remembers to touch is a non-issue.

Install walkthrough with screenshots: Getting Started → Installation.

2026-04-15 — Saddle 4.5.2: Killswitch

The red button is real. One tap in the saddle aborts every in-flight dispatch in the current thread. One tap in the dashboard aborts everything across your entire tenant.

Saddle killswitch (thread-scoped)

• Red “kill” button sits next to “send it.” Aborts every pending or running dispatch in the current thread. Other threads, other workers — untouched. Scalpel, not sledgehammer.
• Works for both single-worker and fan-out dispatches.
• Status bar reports how many jobs were aborted.

Dashboard killswitch (tenant-wide)

• KILL ALL button in the dashboard nav. Two-step confirm before it fires.
• Aborts all in-flight jobs for your tenant and blocks new dispatch until you re-enable.
• Click the button again (now labeled RE-ENABLE) to restore normal operation.

Insurance: per-tenant hourly cap

• A database-level trigger limits how many jobs any tenant can create per hour.
• Enforced in the database itself — no code path can bypass it.
• Added as a defense-in-depth measure so no misconfigured schedule can run away with a budget.

Dashboard security

• Prompt and worker names in the job list are HTML-escaped.
• Browser JS errors are captured for operator review.

---

2026-04-15 — Saddle 4.5.1: Multi-Worker Fan-Out

The headline feature. Dispatch the same prompt to multiple workers at once and compare their answers side by side.

Multi-worker dispatch

• Target picker takes checkboxes. Leave it empty for “router picks.” Tick one to pin a worker. Tick two or more to fan out.
• Side-by-side cards. Each worker’s answer streams into its own card. Read them as they arrive.
• Merged answer. A canonical merged answer appears at the top once everyone’s done — read just that if you’re in a hurry.
• Accept. Tap “accept” on any card to mark it as the one you went with. Tap “copy” for clipboard. The saddle remembers your pick.
• The command strip shows “N workers => fan-out” when you’ve picked two or more.
• Single-worker dispatch is completely unchanged.

---

2026-04-13 — The Monster Rev (4.4.15)

Meet Bob

• Every companion now ships with Bob — a local brain that lives on your machine. Your memories, patterns, and learned behaviors stay with Bob. Not on our servers.
• New accounts are seeded with default routing patterns on first setup.

One-click install

• Download, run, useful in 3 minutes. The wizard handles local routing engine, account creation, worker registration.
• Privacy-first welcome — the first screen explains our data promise.
• Inline signup — create your account directly in the companion.

Context and dispatch

• Workers can check out scoped context from Bob with a time limit and check it back in with results.
• Every piece of data is tagged at creation so the system knows what’s shareable and what’s private.
• Human-readable job IDs.

Worker health

• Workers now have a health lifecycle. Unresponsive workers are automatically removed from dispatch.
• The In Memoriam page honors retired workers.

Security

• Graduated security posture with automatic de-escalation timers. Full shutdown requires human confirmation.

---

2026-04-12 — Companion + Saddle polish

Companion 4.4.12

• Mission Control — the dashboard idle screen shows your live patch: active workers, job history, fleet activity.
• Branded app menu with version display, About screen, documentation link.
• Version visible in tray tooltip.

Companion 4.4.10

• ModelReins logo throughout — window, installer, tray.
• Reliable local model detection (Ollama, LM Studio).
• Brain context injection opt-in and disabled by default.
• Credential pattern detection in the sensitivity filter.

Saddle 4.4.7

• The saddle connects to your patch. Workers, jobs, and routing flow through the patch you own.
• Version is now live in the saddle header.

---

2026-04-11 — The Wall

Companion 4.4.5

• The Wall — windowed mission control with live fleet view, job stream, metrics, leaderboard, and audit trail.
• Page carousel — clickable dots to jump to any page.
• Welcome splash and first-close notice.

Saddle 4.4.5

• Conversation thread — responses render inline, follow-ups carry full context, copy and insert-at-cursor actions.
• Command strip — effort, mode, and target compressed into one clickable line.
• Enter to send, Shift+Enter for new line.

---

2026-04-08 — Cap detection and the saddle picker

Routing, cap detection, and the target picker shipped. Workers now detect Claude session caps in stdout, the router marks capped workers for 60 seconds, and dispatch goes to a different worker automatically.

• Target picker in the saddle — click any worker to pin dispatches to it.
• Local/ollama mode routes through registered ollama-http workers.
• Bring Your Own Harness tour — wrap any agent CLI as a worker.

---

2026-04-04 — v3.1.0: Multi-provider orchestration

• Multi-provider support — Claude, OpenAI, Gemini, Ollama, LM Studio, OpenRouter, 1MinAI, or any CLI tool.
• Provider plugin registry — add new AI tools via YAML config.
• PayPal billing — Free, Pro ($29/mo), Team ($79/mo).
• Multi-tenant brain — engine, tenant, and ops scopes.
• Killswitch — graduated defense levels with automatic de-escalation.
• Zero-knowledge architecture — API keys stay on your machine.
• HMAC-signed audit trail.

---

Earlier releases

For changes before 2026-04-04, see the changelog.