59 agent capabilities. 7 templates. One laptop. Zero tokens.
· Avery NXR
The numbers that describe Avery NXR's agent layer aren't marketing copy. Each one represents a real constraint that other agent platforms accept and we don't. Here's what each number means and why it matters.
59 agent capabilities across 14 categories
What this is: the building blocks you compose to build an agent. Sub-agents. Loops. Conditions. Knowledge base retrieval. File operations. HTTP requests. Shell commands with guardrails. Email and messaging integrations. Database operations. Vector search. Web scraping and rendering. LLM inference. Classification. Extraction. Many more.
Each capability is a node you drop into the agent's graph. The agent is a composition of these capabilities, wired together with the conditional logic that handles the specific workflow.
Why it matters: most agent platforms in 2026 have 15-30 capabilities. The ones that are closer to chatbot platforms have even fewer — they're focused on the conversation, not the work the conversation drives. The ones that are closer to autonomous frameworks have many primitives but require code to compose them.
Avery's 59 capabilities cover the range that operational workflow agents actually need without requiring code. Drag, drop, connect, deploy. The visual builder makes the composition tractable for non-engineers; the YAML escape hatch is there for engineers who want code-level control.
The capabilities aren't novel individually. The number matters because it's the threshold above which most operational workflows can be built without writing custom code or hitting a "we don't support that yet" wall.
7 production-ready agent templates pre-loaded on first launch
What this is: complete working agents that ship with Avery NXR. Anna (daily AI news). Sophia (meeting notes → action items). Marcus (resume screening). Priya (customer support triage). Carlos (daily sales pipeline digest). Yuki (competitor monitoring). Liam (server + endpoint health monitor).
Each one is a real graph you can inspect, run, fork, or modify. Not demos. Not examples in a docs page. Working production agents loaded into the platform on first install.
Why it matters: the gap between "I installed this agent tool" and "I have an agent doing real work" is often a multi-day investment in figuring out what to build and how. The 7 templates close that gap. First-time users can have a real agent processing real work within 10-15 minutes of installation.
The 7 also cover the most common operational use cases across teams of all sizes. Most users find that 3-5 of the 7 templates are directly relevant to their work; they install, configure, run, and they're producing value before they've thought about what custom agents they want to build.
The number 7 isn't magic. It's the count of templates that consistently come up across user types — solo founders, SMBs, agencies, enterprise teams. If we had picked 4, we'd be missing cases. If we'd picked 15, we'd be diluting the curation.
One laptop
What this is: the deployment target. Avery NXR runs on the user's laptop. The model lives in Ollama on the machine. The agent runtime is the desktop application. The connections to external services use OS keychain for credentials.
There is no central server you depend on. No vendor cloud you log into. No remote runtime you have to trust.
Why it matters: every other agent platform assumes cloud deployment as the default. Even the ones with self-hosting options treat self-hosting as a specialized configuration for enterprise customers. The default architecture is "log into our cloud."
Avery flips this. Local is the default. Cloud is an opt-in (Consult Mode) for the rare case that needs frontier reasoning.
The consequences compound: privacy by architecture, cost flat to electricity, latency in milliseconds, sovereignty over your workflows, no vendor cloud to fail.
The "one laptop" framing is literal. The Avery NXR you install on your laptop today will keep working tomorrow whether or not Avery the company exists. You own the runtime.
Zero tokens
What this is: the marginal cost per agent execution. With cloud-LLM-backed agent platforms, every step costs tokens. Every classification, every extraction, every LLM call rings the cash register.
With Avery NXR running on a local model, the marginal cost per execution is zero. There's no token meter. Running an agent 1,000 times today costs the same as running it once.
Why it matters: cloud-LLM agent platforms compound your bill with usage. You can budget for them in isolation, but the bill grows as you do more — which is the opposite of what you want from an automation tool.
Zero marginal cost changes what's possible:
You can run agents continuously instead of selectively. Watch every ticket, every email, every change, every alert.
You can scale workflows you couldn't justify at cloud-LLM prices. Per-customer personalization. Per-rep digests. Per-employee onboarding.
You can experiment without budget overhead. Build an agent to see if it works; if it does, scale it; if it doesn't, retire it. No per-experiment cost.
The economics shift. Operational AI stops being a budget conversation and becomes an "is this worth automating" conversation. The constraint becomes attention, not money.
What the numbers add up to
59 capabilities × 7 templates × 1 laptop × 0 tokens.
The capabilities mean you can build what you need. The templates mean you can start without building. The laptop means it runs without depending on anyone else. The zero tokens means it stays affordable as you scale.
These are four different constraints that other agent platforms accept and we don't. The result is an operational AI platform with different economics, different deployment posture, and different reach than the rest of the category.