Changelog & monthly updates

Flo here — after the v3 paradigm shift in May, June was about giving you the controls.

v3 made Notis dramatically more capable. This release makes that capability yours to dial in: choose exactly how much intelligence each task gets, never hit a wall when you need more, and let Notis run your social calendar end-to-end. Underneath, it was also a heavy month of reliability work to make the new v3 surfaces rock-solid.

Let's get into it.

Intelligence Controls

Most AI agents drown you in knobs — a dozen models, a pile of "effort" settings, endless tradeoffs you were never meant to make. We think that's backwards. So we ran the evals for you and boiled it down to two dials: intelligence and speed. Every other optimization, we made for you.

There are three levels:

• Low — fast and light. The important part: because Low is so much more efficient, the same plan now goes up to 4x further — that's up to 4x more tasks you can delegate to Notis before you hit your usage cap. (Hitting that wall was the single biggest reason people churned — so this one really matters.)

• Medium — the balanced default for everyday work.

• High — our most capable reasoning, for the problems worth thinking hard about.

And on Auto — the default — there's nothing to choose at all. Notis reads each task and picks the right level for you, every time. Prefer to drive? Pro+ users can override intelligence and speed per thread in the apps, per channel, and per automation, right from the Manager.

Learn more about intelligence levels & rates.

On-Demand Usage

Running out of credits mid-task is the worst. So we killed the wall.

With on-demand usage, you can keep going past your plan's limit and simply post-pay for the extra — no upgrade dance, no blocked automation at the exact moment you needed it to fire. You stay in control with usage warnings and clear thresholds, and a new billing overview in your settings shows exactly where you stand.

For teams, credits are now pooled across the whole team instead of being trapped per-seat — so the person who needs them most can actually use them.

Learn more about on-demand usage.

Social Scheduling

Notis can now run your social presence end-to-end — and it's included.

Most social schedulers — Buffer and the others you'll find in the Social Media guide — charge $20–80/month just to queue posts. With Notis it's free, inside the license you already have. We integrated PostForMe natively, so connecting X, LinkedIn, Instagram, TikTok, YouTube and more is a one-click setup in the Integrations tab — no extra tools, no separate subscription.

Then just talk: dictate an idea on your walk, let Notis write and adapt the copy per platform, attach your media, and schedule a whole week of posts from a single message.

Learn more about social media via Notis.

🐞 Other improvements

• Safer sign-in: Strangers can't trigger Notis or quietly create accounts anymore, and signing in by email now uses a simple, secure one-time link.

• Granola memories: Connect Granola and Notis automatically remembers what came up in your meetings (ULTRA).

• Refresh memories on demand: Re-sync your connected memory sources whenever you want, right from the Memories page.

• Smarter, leaner tool use: Notis picks the right tool for the job more reliably — and uses far fewer tokens when you have a lot of integrations connected.

• No more guessing: When an extension isn't connected, Notis now tells you — instead of hallucinating its way through the task you handed off.

• Transcribe anything: Send an audio file and Notis turns it into clean, usable text.

• Better web reading: We upgraded how Notis reads and pulls content from the web, so research comes back cleaner and more complete.

• Inline media in the Manager: Images and files now show up right inside your Manager threads.

• In-app support: A new Contact Support option opens a chat with us without leaving Notis.

• A steadier month: Lots of behind-the-scenes fixes across voice, WhatsApp, billing, and automations so everything from v3 just works.

📰 AI News

Here's what you might have missed in AI since v3:

🔥 Anthropic drops Claude Fable 5 (and its caged twin, Mythos 5) — On June 9 Anthropic shipped Fable 5 publicly while keeping the lightly-safeguarded Mythos 5 to trusted partners. Fable 5 took #1 on FrontierCode Diamond at 29.3% (vs Opus 4.8's 13.4% and GPT-5.5's 5.7%), priced at $10/M input and $50/M output. The system card flags Mythos at a "CB-1" biosafety rating creeping toward CB-2. (Anthropic)

🤖 Claude is now building Claude — Anthropic says 80%+ of production code merged into its codebase is now Claude-authored, with engineers merging 8x more code daily than in 2024. Mythos Preview delivered a 52x model-training optimization speedup and out-predicted human researchers' next move 64% of the time. (Anthropic)

🏛️ Dario Amodei demands binding AI rules — One day after the Fable 5 launch, Anthropic's CEO published a five-part "Policy on the AI Exponential" calling for mandatory FAA-style third-party testing of frontier models, plus government power to block unsafe ones — and pledged "substantial financial backing" to push legislation. (Dario Amodei)

💸 OpenAI eyes an IPO, ships GPT-5.5, and starts a price war — Altman told staff OpenAI expects to go public "within the next year" (SEC paperwork filed), with model 5.6 teed up for June. GPT-5.5 Instant became the ChatGPT default for hundreds of millions, cutting hallucinations 52.5% on high-stakes prompts — as both labs reportedly weigh "drastic" token price cuts ahead of their IPOs. (The Information)

🛡️ Frontier models are getting scarily good at cyber — The UK's AI Security Institute says frontier cyber capability has doubled every 4.7 months since late 2024, with Mythos Preview and GPT-5.5 now blowing past the trend. On the upside, Anthropic's Mythos surfaced 271 Firefox security flaws for Mozilla in two months — same superpower, friendlier hat. (AISI)

🦾 BMW puts humanoids on the line — BMW will deploy Hexagon Robotics' Aeon humanoids (1.65m, 60kg, 21 sensors, 15kg lift) in European manufacturing starting summer 2026 after Leipzig trials. Trained via teleop, NVIDIA digital twins, and RL — the "robots take the factory floor" era is now a press release, not a prediction. (BBC)

🎯 Reality-check benchmarks say agents still aren't there — Two harder benchmarks landed: Workspace-Bench ran 28 agent setups on 388 real office tasks (best: Opus 4.7 + OpenClaw at 68.7% vs an 80.7% human baseline), and DeepSWE's 113 tasks average 668 lines of code per fix. Translation: agents are improving fast but still trail humans on messy real work — exactly the gap tools like Notis live in. (DeepSWE)

That's it for June. v3 gave Notis its new shape; this month makes it yours to tune.

See you next month. Flo

P.S.: Record a video testimonial (in any language) and get a free month of Notis on us.