For two years the trimio product talked about itself in borrowed vocabulary: routing rules, LCR configuration, least cost logic. The territory was ours. The language was everyone else's.
As of the proxy release this week, that's over.
The unit of work in trimio is now a trimio token — a routing token the system creates, adjusts, and routes on your behalf. Writing a trimio token used to mean writing a routing rule by hand. Today it means describing routing intent in a sentence and letting the trimio rule generator build the token for you.
Three things changed simultaneously and they matter together:
The change is not a rebrand alone. It is a deliberate adoption-floor rewrite. This post is about what shipped, why those three things ship together, and what trimio tokens mean for the way enterprise AI routing is talked about going forward.
A noun the customer uses is a noun that survives the sales conversation. "I added a routing rule to fall back on Claude Haiku when Sonnet is rate-limited" is a sentence you only finish if you understood what you clicked. "I added a trimio token with a Haiku fallback on the Sonnet rate limit" is a sentence you finish in a board memo.
The same idea explains how Stripe landed PaymentIntent in 2017: they did not invent a new category, they named a primitive inside an existing one. A buyer who says "I implemented PaymentIntents" has internalized the category. A buyer who says "I configured the new payment flow" has not.
trimio tokens are the routing primitive. buyers no longer say "I configured an LCR rule." They say "I have trimio tokens covering tier A, tier B, and tier C." That sentence is the category, the product, and the budget line in eight words.
Six PRs in this week's LCR sprint:
Anyone who has rolled out multi-provider routing at enterprise scale knows the first week is the rough one. The team wants to route calls away from the most expensive provider when a cheaper model meets the quality bar. They have model IDs from a provider they have never used. They have a quality threshold expressed differently by every provider. They have to reason about retries, fallbacks, and cache hits across endpoints.
The previous solution was to write a routing rule by hand. That is what every gateway in the category asks you to do, and it works for the engineer who built it. It does not survive that engineer going on vacation.
The rule generator inverts the onboarding question. Instead of "do you know how to configure a router?" we ask "describe how you want to route." The system handles the model IDs, the quality floor math, the fallback ordering, and the platform-vs-org scoping automatically.
Concretely, the prompt that previously required a YAML block now looks like this:
"Default to Sonnet 5 for coding questions because the team has learned its style. If Sonnet is rate-limited, fall through to GPT-5.6 Terra because it's ~5× cheaper and the quality is acceptable for our grading rubric. Never route our compliance workload off Claude. If we hit a hard cap on any provider, surface an alert to the #ai-ops Slack channel."
A trimio token is generated from that sentence in seconds. The user reviews and edits it before saving. What used to be a 30-minute exercise with reference docs becomes a 10-second authoring step the finance team's first-week analyst can run.
PR #1019 (per-tenant tier presets) is the change customers will notice before they ever see the rule generator. The presets ship three tier defaults:
Tenants land in trimio, see Tier B as the default, and start routing. If they want different policy, they describe the change to the rule generator. The token is rebuilt against the merge config facade. The merged config view in the UI shows all of their tokens with scope annotations so the platform-vs-org boundary stays clear.
Before the merged config facade shipped, customers saw two views: platform rules (which they could not always edit) and org rules (which they could). The plumbing fused them but the UI presented them in two places. That gap is where enterprise procurement pushes back: "I don't know who set which rule, where, or when."
The merged config facade merges both surfaces into one editable view, with clear visual markup showing which tokens are platform-scope (managed, immutable from inside the org) and which are org-scope (customer-authored, change-controlled by the org admin). The same facade is what the rule generator writes against, which means the AI-generated tokens appear in the merged config alongside human-authored ones.
The enterprise SOC 2 argument reads cleanly: every routing decision has a tenant-scoped audit trail, every change is visible in one place, every token carries a scope annotation that survives every screenshot taken for compliance review. That is rare in this category.
The nearest comparable features in the category today, in production:
The combination trimio is now shipping — natural-language rule generation, per-tenant tier presets, a merged config facade, role-gating on both sides of the platform-vs-org boundary, and a live Release Notes section showing every shift — does not exist elsewhere in the category. The vocabulary move (trimio tokens) gives the engineering org a noun to use in sales calls and procurement paperwork that no competitor owns.
Two follow-ups are worth watching.
First, the production cutover. trimio-demo is at v0.7.1120 with the rule generator and tokens rebrand live. proxy-test is at v0.7.1106, which is the prior sprint. The LCR sprint promotion to production is a single trimio-infra PR pinning proxy-test to v0.7.1120. Until that lands, the trimio tokens story can be told but not yet shown end-to-end in customer environments on the prior version. The two-week window before the next enterprise demo is the window this should land in.
Second, vocabulary propagation. The rebrand is a noun in the product; it is not yet a noun in the sales deck. Every demo walkthrough, every procurement response, every FAQ entry needs to use trimio tokens within Q3 for the rebrand to compound. That means updating sales enablement, customer-facing docs, the platform pitch, the product page, the on-call escalation doc, and the SOC 2 control narrative. None of that is technical work; all of it is roll-out work.
This week's LCR sprint did three things simultaneously: it named the unit of work (trimio tokens), it let the gateway generate that unit from a sentence (the trimio rule generator), and it gave every new tenant usable policy from the first session (per-tenant tier presets). The merged config facade and role-gating are not feature additions — they are the boundaries that make the AI-generated tokens safe for enterprise procurement.
For a category that has been configuring routing rules by hand since 2023, this is a step-change. The category caught up. The vocabulary gets adopted. The question for downstream is whether to copy the rule generator, copy the tokens, or copy both.
trimio is the LLM API gateway built for AI cost governance — multi-provider routing, AI-native rule authoring, content compression, and CFO-grade audit trails in one layer. See how trimio tokens work in production.