diff --git a/AGENTS.md b/AGENTS.md
index f33bdfb..f499554 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -5,7 +5,7 @@ Guidance for agents working in this repository.
 ## What this is
 
 An OpenCode V2 plugin (`context-limit.ts`) that sets a per-model working context
-budget by lowering the model's `limit.context` through a catalog transform. No
+budget by lowering the model's `limit.context` through a model transform. No
 build step, no dependencies, MIT.
 
 ## Local development
@@ -24,9 +24,9 @@ grep context-limit ~/.local/share/opencode/log/opencode.log | tail
 
 ## Spike result (T0)
 
-A catalog transform can lower a model's context window at runtime. A probe set
+A model transform can lower a model's context window at runtime. A probe set
 `deepseek/deepseek-flash` from 1,000,000 to 123,456 through
-`ctx.catalog.transform`, and a re-read showed 123,456. Compaction's default
+`ctx.model.transform`, and a re-read showed 123,456. Compaction's default
 threshold follows the model's usable input budget, so this is the mechanism. No
 config edit is needed.
 
@@ -40,9 +40,9 @@ config edit is needed.
 
 ## API notes
 
-- `ctx.catalog.transform((catalog) => catalog.model.update(providerID, modelID, (model) => { model.limit = { ...model.limit, context: n } }))`
-  lowers a window. Call `ctx.catalog.reload()` after changing the rules.
-- Model entries from `ctx.catalog.model.list()` carry `providerID`, `id`, and
+- `ctx.model.transform((editor) => editor.update(providerID, modelID, (model) => { model.limit = { ...model.limit, context: n } }))`
+  lowers a window. Call `ctx.model.reload()` after changing the rules.
+- Model entries from `ctx.model.list()` carry `providerID`, `id`, and
   `limit.context`.
 - Rules live in `ctx.storage` under `context-limit`.
 
@@ -50,9 +50,9 @@ config edit is needed.
 
 - `parseBudget` - parses tokens, `128K`, `1M`, and `50%`, with clamping.
 - `matchPattern`, `longestMatch`, `resolveBudget` - rule matching.
-- `applyBudget` - the catalog transform body, exported for tests.
+- `applyBudget` - the model transform body, exported for tests.
 - `setup` - registers the command and the transform.
-- `context-limit.test.ts` - tests with a fake catalog and ctx.
+- `context-limit.test.ts` - tests with a fake model editor and ctx.
 
 ## Releasing
 
diff --git a/README.md b/README.md
index 5881c0f..de10f88 100644
--- a/README.md
+++ b/README.md
@@ -43,9 +43,9 @@ matches everything. The longest matching pattern wins.
 
 ## How it works
 
-The plugin registers a catalog transform that lowers the matched models'
+The plugin registers a model transform that lowers the matched models'
 `limit.context`. Compaction's default threshold follows the model's usable input
-budget, so compaction fires earlier. A change calls `ctx.catalog.reload()` and
+budget, so compaction fires earlier. A change calls `ctx.model.reload()` and
 applies at once. Nothing is written to `opencode.json`.
 
 ## Tests
diff --git a/context-limit.test.ts b/context-limit.test.ts
index 16b6868..5d3c732 100644
--- a/context-limit.test.ts
+++ b/context-limit.test.ts
@@ -1,28 +1,21 @@
 import { describe, expect, test } from "bun:test"
 import plugin, { applyBudget, longestMatch, matchPattern, parseBudget, resolveBudget, VERSION } from "./context-limit.ts"
 
-function makeCatalog(models: Array<{ providerID: string; id: string; context: number }>) {
+function makeEditor(models: Array<{ providerID: string; id: string; context: number }>) {
   const entries = models.map((model) => ({
     providerID: model.providerID,
     id: model.id,
     limit: { context: model.context, output: 1000 },
   }))
-  const providers = new Map<string, { providerID: string; models: Map<string, any> }>()
-  for (const entry of entries) {
-    if (!providers.has(entry.providerID)) providers.set(entry.providerID, { providerID: entry.providerID, models: new Map() })
-    providers.get(entry.providerID)!.models.set(entry.id, entry)
-  }
-  const catalog: any = {
-    provider: { list: () => [...providers.values()] },
-    model: {
-      update: (providerID: string, modelID: string, change: (model: any) => void) => {
-        const model = providers.get(providerID)?.models.get(modelID)
-        if (model) change(model)
-      },
+  const editor: any = {
+    list: () => entries,
+    update: (providerID: string, modelID: string, change: (model: any) => void) => {
+      const model = entries.find((entry) => entry.providerID === providerID && entry.id === modelID)
+      if (model) change(model)
     },
     entries,
   }
-  return catalog
+  return editor
 }
 
 function makeCtx(options: { model?: any; models?: any[] } = {}) {
@@ -36,22 +29,15 @@ function makeCtx(options: { model?: any; models?: any[] } = {}) {
     ]
   ).map((model) => structuredClone(model))
 
-  // Rebuild the catalog from the registered transforms on every read, the way
-  // the runtime replays transforms onto a fresh value.
+  // Rebuild the model list from the registered transforms on every read, the
+  // way the runtime replays transforms onto a fresh value.
   const rebuild = () => {
     const entries = structuredClone(baseModels)
-    const providers = new Map<string, { providerID: string; models: Map<string, any> }>()
-    for (const entry of entries) {
-      if (!providers.has(entry.providerID)) providers.set(entry.providerID, { providerID: entry.providerID, models: new Map() })
-      providers.get(entry.providerID)!.models.set(entry.id, entry)
-    }
     const editor = {
-      provider: { list: () => [...providers.values()] },
-      model: {
-        update: (providerID: string, modelID: string, change: (model: any) => void) => {
-          const model = providers.get(providerID)?.models.get(modelID)
-          if (model) change(model)
-        },
+      list: () => entries,
+      update: (providerID: string, modelID: string, change: (model: any) => void) => {
+        const model = entries.find((entry) => entry.providerID === providerID && entry.id === modelID)
+        if (model) change(model)
       },
     }
     for (const transform of transforms) transform(editor)
@@ -66,10 +52,10 @@ function makeCtx(options: { model?: any; models?: any[] } = {}) {
     session: {
       get: async () => ({ model: options.model ?? { providerID: "opencode-go", id: "deepseek-v4.1-flash" } }),
     },
-    catalog: {
+    model: {
       transform: async (callback: any) => void transforms.push(callback),
       reload: async () => void (reloads += 1),
-      model: { list: async () => ({ data: rebuild() }) },
+      list: async () => ({ data: rebuild() }),
     },
     command: { transform: (callback: any) => callback({ add: (definition: any) => commands.push(definition) }) },
   }
@@ -128,20 +114,20 @@ describe("longestMatch and resolveBudget", () => {
 
 describe("applyBudget", () => {
   test("lowers only matched models", () => {
-    const catalog = makeCatalog([
+    const editor = makeEditor([
       { providerID: "opencode-go", id: "x", context: 1_000_000 },
       { providerID: "deepseek", id: "y", context: 1_000_000 },
     ])
-    applyBudget(catalog, [{ pattern: "opencode-go/*", value: 128_000, unit: "tokens" }])
-    const byKey = Object.fromEntries(catalog.entries.map((entry: any) => [`${entry.providerID}/${entry.id}`, entry.limit.context]))
+    applyBudget(editor, [{ pattern: "opencode-go/*", value: 128_000, unit: "tokens" }])
+    const byKey = Object.fromEntries(editor.entries.map((entry: any) => [`${entry.providerID}/${entry.id}`, entry.limit.context]))
     expect(byKey["opencode-go/x"]).toBe(128_000)
     expect(byKey["deepseek/y"]).toBe(1_000_000)
   })
 
   test("does nothing without rules", () => {
-    const catalog = makeCatalog([{ providerID: "a", id: "b", context: 1000 }])
-    applyBudget(catalog, [])
-    expect(catalog.entries[0].limit.context).toBe(1000)
+    const editor = makeEditor([{ providerID: "a", id: "b", context: 1000 }])
+    applyBudget(editor, [])
+    expect(editor.entries[0].limit.context).toBe(1000)
   })
 })
 
@@ -208,9 +194,9 @@ describe("setup", () => {
     expect(commands.map((entry) => entry.name)).toEqual(["context-limit"])
     expect(transforms).toHaveLength(1)
 
-    const catalog = makeCatalog([{ providerID: "opencode-go", id: "x", context: 1_000_000 }])
-    transforms[0](catalog)
-    expect(catalog.entries[0].limit.context).toBe(128_000)
+    const editor = makeEditor([{ providerID: "opencode-go", id: "x", context: 1_000_000 }])
+    transforms[0](editor)
+    expect(editor.entries[0].limit.context).toBe(128_000)
   })
 })
 
diff --git a/context-limit.ts b/context-limit.ts
index 982dce5..4fa6dbe 100644
--- a/context-limit.ts
+++ b/context-limit.ts
@@ -1,14 +1,14 @@
 // OpenCode V2 context-limit plugin.
 //
 // Sets a working context budget per model by lowering the model's
-// `limit.context` through a catalog transform. Compaction's default threshold
+// `limit.context` through a model transform. Compaction's default threshold
 // follows the model's usable input budget, so a lower window makes compaction
 // fire earlier. A budget can only lower a window, never raise it.
 //
 // The runtime does not resolve @opencode/plugin, so this file exports a plain
 // { id, setup } object.
 
-const VERSION = "0.1.2"
+const VERSION = "0.1.3"
 
 type Unit = "tokens" | "percent"
 
@@ -60,19 +60,17 @@ function resolveBudget(rules: Rule[], key: string, catalogContext: number): numb
   return rule ? budgetFor(rule, catalogContext) : undefined
 }
 
-// The catalog transform body. `catalog` is a catalog editor: it exposes
-// provider.list() and model.update(providerID, modelID, change).
-function applyBudget(catalog: any, rules: Rule[]): void {
+// The model transform body. `editor` is a model editor: it exposes
+// list() and update(providerID, modelID, change).
+function applyBudget(editor: any, rules: Rule[]): void {
   if (rules.length === 0) return
-  for (const record of catalog.provider.list()) {
-    for (const model of record.models.values()) {
-      const key = `${model.providerID}/${model.id}`
-      const budget = resolveBudget(rules, key, model.limit.context)
-      if (budget === undefined) continue
-      catalog.model.update(model.providerID, model.id, (entry: any) => {
-        entry.limit = { ...entry.limit, context: budget }
-      })
-    }
+  for (const model of editor.list()) {
+    const key = `${model.providerID}/${model.id}`
+    const budget = resolveBudget(rules, key, model.limit.context)
+    if (budget === undefined) continue
+    editor.update(model.providerID, model.id, (entry: any) => {
+      entry.limit = { ...entry.limit, context: budget }
+    })
   }
 }
 
@@ -86,7 +84,7 @@ async function saveRules(ctx: any, rules: Rule[]): Promise<void> {
 }
 
 async function modelContext(ctx: any, key: string): Promise<number | undefined> {
-  const result = await ctx.catalog.model.list()
+  const result = await ctx.model.list()
   const data: any[] = Array.isArray(result) ? result : (result?.data ?? [])
   const found = data.find((model) => `${model.providerID}/${model.id}` === key)
   return found?.limit?.context
@@ -104,7 +102,7 @@ const plugin = {
   async setup(ctx: any) {
     let rules = await loadRules(ctx)
 
-    await ctx.catalog.transform((catalog: any) => applyBudget(catalog, rules))
+    await ctx.model.transform((editor: any) => applyBudget(editor, rules))
 
     await ctx.command.transform((editor: any) => {
       editor.add({
@@ -163,7 +161,7 @@ const plugin = {
           if (clears) {
             rules = rules.filter((rule) => rule.pattern !== pattern)
             await saveRules(ctx, rules)
-            await ctx.catalog.reload()
+            await ctx.model.reload()
             return
           }
           if (!parsed || (parsed.unit === "tokens" && parsed.value <= 0)) {
@@ -172,7 +170,7 @@ const plugin = {
           rules = rules.filter((rule) => rule.pattern !== pattern)
           rules.push({ pattern, value: parsed.value, unit: parsed.unit })
           await saveRules(ctx, rules)
-          await ctx.catalog.reload()
+          await ctx.model.reload()
         },
       })
     })
diff --git a/docs/compose/spec/context-limit.md b/docs/compose/spec/context-limit.md
index a661b48..dc8da2c 100644
--- a/docs/compose/spec/context-limit.md
+++ b/docs/compose/spec/context-limit.md
@@ -10,11 +10,11 @@ commits: 5e865ff..4647039
 
 ## Report
 
-**T0 spike result** - The mechanism is a catalog transform:
+**T0 spike result** - The mechanism is a model transform:
 
 ```ts
-ctx.catalog.transform((catalog) =>
-  catalog.model.update(providerID, modelID, (model) => {
+ctx.model.transform((editor) =>
+  editor.update(providerID, modelID, (model) => {
     model.limit = { ...model.limit, context: n }
   }),
 )
@@ -27,9 +27,9 @@ usable input budget, so this is the mechanism, and no config edit is needed.
 **What was built** - A single-file OpenCode V2 plugin that sets a working
 context budget per model. `/context-limit` shows the budget, `/context-limit
 128K` or `50%` sets it for the current model, `/context-limit <pattern>
-<value>` sets a rule, and `0` clears. Rules live in storage. A catalog transform
+<value>` sets a rule, and `0` clears. Rules live in storage. A model transform
 lowers the matched models' `limit.context`, and a change calls
-`ctx.catalog.reload()`. Budgets clamp to the catalog window and never raise it.
+`ctx.model.reload()`. Budgets clamp to the catalog window and never raise it.
 
 **Verification** - `bun test`: 14 pass, 0 fail, 42 assertions. Live: `128K`
 lowered the effective window to 128000; `50%` reported window 500000 with
@@ -38,14 +38,14 @@ blocking items plus a medium and lows; all are resolved.
 
 **Journey log**
 
-1. The spike proved the mechanism: a catalog transform lowered
+1. The spike proved the mechanism: a model transform lowered
    `deepseek/deepseek-flash` from 1,000,000 to 123,456, so no config edit was
    needed.
 2. The show path re-resolved a percent rule against the already-lowered window,
    so a 50% rule printed Budget 250000. It now reports the stored rule, and a
    token rule above the window prints the clamped number.
-3. The show test used a fixed catalog, which hid that bug. The fake now rebuilds
-   the catalog from the registered transforms, the way the runtime replays them.
+3. The show test used a fixed model list, which hid that bug. The fake now rebuilds
+   the list from the registered transforms, the way the runtime replays them.
 
 ## [S1] Problem
 
@@ -68,7 +68,7 @@ compaction.
   such as `opencode-go/*` or `*`, and the longest matching pattern wins. The map
   lives in `opencode.json` under `compaction.max_context` when the config
   supports it, otherwise in plugin storage.
-- A spike task decides where the budget can take effect: a catalog transform on
+- A spike task decides where the budget can take effect: a model transform on
   the model limit, or the config compaction threshold. The chosen path is
   recorded in the spec before the command is built.
 
diff --git a/package.json b/package.json
index 1ce1689..0c1f5f9 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "opencode-context-limit",
-  "version": "0.1.2",
+  "version": "0.1.3",
   "description": "OpenCode V2 plugin that sets a per-model working context budget",
   "type": "module",
   "license": "MIT",