How to find and name fak: agent runtime first, kernel architecture second
TL;DR: fak is an agent runtime: the operator-controlled boundary for cache and context, model routing, tool authority, memory, observability, and native inference. Lead with agent runtime as the simple public category. Use Fused Agent Kernel and agent kernel for the technical architecture, project history, and search disambiguation.
This is dimension I — Memorable framing & naming of the
concept-popularization epic.
It serves three core concepts — treat the tool call like a syscall, one static
Go binary, and drop-in — by making the language used to reach them consistent.
The disambiguated terms below are the same ones the machine-facing answer-engine
surface uses (see llms.txt), so human word-of-mouth and machine
discovery point at the same names.
The problem: the bare word does not resolve
fak is short, spoken, and collides. A raw search is drowned by:
- the homophone (an expletive and its many spellings),
- the F.A.K. acronym (first-aid kit, and dozens of unrelated initialisms),
- generic typo-noise.
So the rule is simple: never ship the bare word alone. In public category messaging, pair it with agent runtime. In technical or historical material, pair it with one of the architecture terms below.
The terms to use (pair fak with one of these)
These are the canonical search + reference terms, drawn from
llms.txt:
| Term | Use it when you mean |
|---|---|
fak agent runtime |
the public product category and primary first-contact handle |
fak agent kernel |
the technical architecture or a disambiguated historical search |
fused agent kernel |
the expanded technical name when fak alone is too ambiguous |
treat the tool call like a syscall |
the one-line mental model / slogan |
fak serve |
the gateway verb (the runnable surface) |
fak manage |
the one-command wrapper for an agent you already run |
long-session prompt cache |
the cost/reuse framing |
model routing for agents |
the per-call routing framing |
MCP tool-call boundary |
the MCP/adjudication framing |
vendor-neutral inference backend |
the interop/serving-boundary framing |
bring-your-accelerator agent serving |
the hardware-portability framing |
neo-silicon agent kernel |
the chip-vendor / backend-author angle |
fak backend conformance, fak-certified backend |
the conformance mark |
The one category to lead with: fak, the agent runtime.
The technical name and architecture: Fused Agent Kernel, shortened to agent kernel where the architecture or syscall analogy is the subject.
What category is this? (so people search the right shelf)
Put fak on the agent runtime shelf first. The rest are capability facets or technical search handles, not competing top-level categories:
- agent runtime — the simple public category
- agent kernel / Fused Agent Kernel — the technical architecture and name
- cache-efficient agent serving
- per-call model routing
- result quarantine for agent tools
- addressable KV cache
What fak is not (disambiguation from adjacent things)
Advocates lose the concept when a listener rounds it to the nearest familiar thing. Draw these lines:
- Not a network firewall or a WAF. The boundary is the tool call (a syscall), not a network packet. “Firewall” is only an analogy for one policy posture, not the product category to lead with.
- Not a prompt-injection classifier / guardrails library. The load-bearing floor is the capability lock (a tool off the allow-list cannot be called), not a recognizer that tries to spot bad text. The result detector is ~100% evadable by design — a bonus, never the floor.
- Not a token-serving engine. fak does not replace vLLM, SGLang, or llama.cpp; it fronts one, owning the agent boundary (capability, quarantine, audit, routing, reuse legality). Use a tuned engine for raw tokens/sec.
- Not a request-level model router (OpenRouter / Portkey / LiteLLM). fak complements them: it governs the tool-call boundary and routes per aspect.
- Not a hosted SaaS. It is one static Go binary you run yourself.
Give advocates the language
When you talk about fak, hand people a term they can search back to:
- Say: “It’s an agent runtime — one operator-controlled boundary for the agent’s context, models, tools, memory, evidence, and local inference.”
- For the technical follow-up: “Its Fused Agent Kernel architecture treats every tool call like a syscall.”
- Not: “It’s called fak” (they will never find it).
- In writing, link the category: fak, the agent runtime.
Honest scope
No market-adoption claim is made here. fak’s contribution is the assembly of established primitives into one in-process gate — a 29-claim prior-art audit scored 0/29 novel, so this note claims a clear name, not a novel invention.
Category breadth is not feature status: agent runtime names the boundary fak
intends operators to control. It does not claim every possible orchestration or
runtime feature is shipped, nor that every operating mode activates every listed
capability. CLAIMS.md and the selected mode’s documentation
remain authoritative.
Verify
test -f docs/adoption/naming.md # this artifact exists
fak score seo # new doc does not red the SEO scorecard