fix(devin): resolve models through catalog family metadata

docs-site/src/content/docs/reference/adapters.md

@@ -571,8 +571,10 @@ configuration that names the old id is rewritten at startup.
   session token doubled and dash-joined in an `Authorization: Basic` header while the protobuf body
   keeps one copy, the request envelope goes up uncompressed, and `Metadata` #31 carries a
   732-character device fingerprint whose length — not value — the service checks. Inside
-  `CompletionConfiguration`, #2 is the output cap and #3 is the context window; swapping those two
-  makes every turn fail with an opaque `invalid_argument`. A temperature of exactly 0 is refused, so
+  `CompletionConfiguration`, #2 is the output cap and #3 is `max_newlines`, sent at a fixed large
+  value; swapping those two makes every turn fail with an opaque `invalid_argument`. With no caller
+  or configured cap, the output cap is the selected model's own ceiling from the catalog, and 8192
+  only when the catalog is unavailable. A temperature of exactly 0 is refused, so
   it is clamped to the smallest accepted value.
 - A pre-output 429 with a stated recovery delay is surfaced immediately by default, releasing the
   admitted turn's shared capacity. Set `OPENCODEX_DEVIN_STATED_RESET_WAIT_MS` to a positive cumulative
@@ -591,13 +593,18 @@ configuration that names the old id is rewritten at startup.
 - Experimental unofficial bridge; not shown in the dashboard preset by default. See the
   [provider guide](/guides/providers/) for login instructions.
 
-For SWE-2, an explicit reasoning effort overrides an effort suffix in the model
-id. For example, `swe-2-high` with `medium` selects the native `swe-2-medium` UID;
-`xhigh`, `ultra`, and `max` select `swe-2-max`. Values below Medium select Medium
-and do not disable SWE-2 reasoning. Without an explicit effort, a suffixed model
-id is preserved. This applies through the shared adapter to every Devin account,
-whichever login path minted the credential; other model families keep their
-existing suffix precedence.
+The model id you pick is a model family, and the reasoning effort picks the
+variant. With no effort, the family's own default variant is used: `swe-1-7`
+selects `swe-1-7-medium` and `swe-2` selects `swe-2-high`. An effort changes only
+the effort and lands on the lowest variant at or above it, or the highest one
+below when nothing is above, so a missing rung never turns reasoning down:
+`swe-1-7` at `high` selects the Max row `swe-1-7`, `swe-2` at `low` selects
+`swe-2-medium`, and `kimi-k3` at `medium` selects `kimi-k3-high`. `fast` selects the Fast variant and `1m`, `max-1m` or `none-1m`
+the 1M-context variant where the family has one; otherwise they change nothing.
+A suffixed id such as `claude-opus-5-high-fast` keeps its variant without an
+effort, and with one keeps its Fast and context settings. Resolution never moves
+to another family. When the account catalog is unavailable, the effort is
+appended to the id instead.
 
 ## `azure-openai` (alias: `azure`)
 

scripts/test-layout/layout.json

@@ -911,6 +911,7 @@
     "devin-adapter.test.ts": "providers",
     "devin-cli-authmode-migration.test.ts": "providers",
     "devin-effort-ladder.test.ts": "providers",
+    "devin-family-resolution.test.ts": "providers",
     "devin-hardening.test.ts": "providers",
     "devin-image-passthrough.test.ts": "providers",
     "devin-live-models.test.ts": "providers",

src/adapters/devin/cloud-direct/catalog.ts

@@ -35,8 +35,18 @@
  *     #5  supports_images      bool   ← tri-state: absent stays unknown
  *     #18 max_input_tokens     varint ← per-account context window
  *     #22 model_uid            string ← what `GetChatMessage` accepts
+ *     #23 model_info           ModelInfo { #6 features { #15 supports_thinking },
+ *                                          #13 max_output_tokens, #23 model_family_uid }
+ *     #30 family_metadata      ModelFamilyMetadata { #1 family_label,
+ *                                          #2 repeated Entry { #1 axis key, #2 { #1 order, #2 name } },
+ *                                          #3 is_default_model_in_family }
+ *     #31 is_default_model_in_family bool
  *   }
  *
+ *   #23/#30/#31 were read back from a live 267-row catalog (2026-09-27): every
+ *   current row names its family, and each family marks one default member.
+ *   Legacy `MODEL_*` rows carry no #30.
+ *
  *   Disabled semantics: TRUE means "this UID exists in the catalog but the
  *   caller's account/tier cannot run inference against it." BYOK models
  *   surface as `disabled: false` so users with their own provider keys still
@@ -91,6 +101,73 @@ export interface ModelCatalogEntry {
    * (see src/providers/antigravity-models.ts).
    */
   supportsImages?: boolean;
+  /** ModelInfo #13: the model's own output-token ceiling. */
+  maxOutputTokens?: number;
+  /** ModelInfo #6.15. */
+  supportsThinking?: boolean;
+  /** ModelInfo #23, e.g. `swe-1.7` for both `swe-1-7` and `swe-1-7-medium`. */
+  familyUid?: string;
+  /** ModelFamilyMetadata #1, e.g. `SWE-1.7`. */
+  familyLabel?: string;
+  /**
+   * ModelFamilyMetadata #2: this row's position on each of its family's axes,
+   * keyed by axis name (`Effort`, `Reasoning Effort`, `Fast Mode`, `1M Context`,
+   * `Thinking`, ...). Toggle axes such as `Fast Mode` carry only an order.
+   */
+  familyAxes?: Record<string, DevinFamilyAxisValue>;
+  /** #31 (or #30.3): the member the family selects when nothing is asked for. */
+  isFamilyDefault?: boolean;
+}
+
+export interface DevinFamilyAxisValue {
+  order: number;
+  name?: string;
+}
+
+/**
+ * Rows whose catalog entry claims image support that the model does not have.
+ * Live 2026-09-27: `swe-1-6` answered "NOIMAGE" to a solid red PNG while
+ * `swe-1-6-fast` and `swe-2-medium` named the colour, so the image is dropped
+ * server-side without an error. Marking it text-only lets the vision fallback
+ * describe the image instead of the model silently never seeing it.
+ */
+const IMAGE_BLIND_UIDS = new Set(['swe-1-6']);
+
+function parseFamilyMetadata(buf: Buffer): { label?: string; axes: Record<string, DevinFamilyAxisValue>; isDefault: boolean } {
+  let label: string | undefined;
+  let isDefault = false;
+  const axes: Record<string, DevinFamilyAxisValue> = {};
+  for (const f of iterFields(buf)) {
+    if (f.num === 1 && f.wire === 2 && Buffer.isBuffer(f.value)) label = f.value.toString('utf8');
+    else if (f.num === 3 && f.wire === 0) isDefault = f.value === 1n;
+    else if (f.num === 2 && f.wire === 2 && Buffer.isBuffer(f.value)) {
+      let key = '';
+      const value: DevinFamilyAxisValue = { order: 0 };
+      for (const e of iterFields(f.value)) {
+        if (e.num === 1 && e.wire === 2 && Buffer.isBuffer(e.value)) key = e.value.toString('utf8');
+        else if (e.num === 2 && e.wire === 2 && Buffer.isBuffer(e.value)) {
+          for (const v of iterFields(e.value)) {
+            if (v.num === 1 && v.wire === 0) value.order = Number(v.value);
+            else if (v.num === 2 && v.wire === 2 && Buffer.isBuffer(v.value)) value.name = v.value.toString('utf8');
+          }
+        }
+      }
+      if (key) axes[key] = value;
+    }
+  }
+  return { ...(label ? { label } : {}), axes, isDefault };
+}
+
+function parseModelInfo(buf: Buffer): { maxOutputTokens?: number; familyUid?: string; supportsThinking?: boolean } {
+  const out: { maxOutputTokens?: number; familyUid?: string; supportsThinking?: boolean } = {};
+  for (const f of iterFields(buf)) {
+    if (f.num === 13 && typeof f.value === 'bigint' && f.value > 0n) out.maxOutputTokens = Number(f.value);
+    else if (f.num === 23 && f.wire === 2 && Buffer.isBuffer(f.value) && f.value.length > 0) out.familyUid = f.value.toString('utf8');
+    else if (f.num === 6 && f.wire === 2 && Buffer.isBuffer(f.value)) {
+      for (const g of iterFields(f.value)) if (g.num === 15 && g.wire === 0) out.supportsThinking = g.value === 1n;
+    }
+  }
+  return out;
 }
 
 export interface CacheEntry {
@@ -127,6 +204,9 @@ export function parseCatalogBuffer(buf: Buffer, apiKey: string, host: string): C
     let disabled = false;
     let contextWindow = 0;
     let supportsImages: boolean | undefined;
+    let info: ReturnType<typeof parseModelInfo> = {};
+    let family: ReturnType<typeof parseFamilyMetadata> | undefined;
+    let isFamilyDefault = false;
     for (const sf of iterFields(f.value as Buffer)) {
       if (sf.num === 1 && sf.wire === 2 && Buffer.isBuffer(sf.value)) {
         label = (sf.value as Buffer).toString('utf8');
@@ -145,15 +225,26 @@ export function parseCatalogBuffer(buf: Buffer, apiKey: string, host: string): C
         contextWindow = Number(sf.value);
       } else if (sf.num === 22 && sf.wire === 2 && Buffer.isBuffer(sf.value)) {
         modelUid = (sf.value as Buffer).toString('utf8');
+      } else if (sf.num === 23 && sf.wire === 2 && Buffer.isBuffer(sf.value)) {
+        info = parseModelInfo(sf.value);
+      } else if (sf.num === 30 && sf.wire === 2 && Buffer.isBuffer(sf.value)) {
+        family = parseFamilyMetadata(sf.value);
+      } else if (sf.num === 31 && sf.wire === 0) {
+        isFamilyDefault = sf.value === 1n;
       }
     }
+    if (IMAGE_BLIND_UIDS.has(modelUid)) supportsImages = false;
     if (modelUid.length > 0) {
       byUid.set(modelUid, {
         modelUid,
         label: label || modelUid,
         disabled,
         ...(contextWindow > 0 ? { contextWindow } : {}),
         ...(supportsImages !== undefined ? { supportsImages } : {}),
+        ...info,
+        ...(family?.label ? { familyLabel: family.label } : {}),
+        ...(family ? { familyAxes: family.axes } : {}),
+        ...(isFamilyDefault || family?.isDefault ? { isFamilyDefault: true } : {}),
       });
     }
   }

src/adapters/devin/cloud-direct/chat.ts

@@ -332,10 +332,15 @@ function collapseSystemIntoUser(messages: ChatHistoryItem[]): ChatHistoryItem[]
  * CompletionConfiguration — mirrors the LS-shipped defaults, lets the caller
  * override the obvious knobs.
  */
-/** Output cap when the caller named none. */
+/** Output cap when neither the caller nor the catalog named one. */
 const DEFAULT_MAX_OUTPUT_TOKENS = 8192;
-/** Context window when the caller named none. */
-const DEFAULT_CONTEXT_WINDOW = 128_000;
+/**
+ * CompletionConfiguration #3 is `max_newlines`, not a token count and not an
+ * input ceiling: live, a value of 5 did not truncate a 25-line answer. It is
+ * still sent, at the value every turn has carried, so the request shape the
+ * service accepts does not change.
+ */
+const MAX_NEWLINES = 128_000;
 
 /**
  * Cognition rejects a temperature of exactly 0 with the same opaque internal
@@ -353,7 +358,6 @@ function safeTemperature(value: number | undefined): number {
 
 function encodeCompletionConfiguration(opts: {
   maxOutputTokens?: number;
-  maxInputTokens?: number;
   temperature?: number;
   topK?: number;
   topP?: number;
@@ -365,15 +369,15 @@ function encodeCompletionConfiguration(opts: {
   };
   // Tag map, verified by building the same turn with a working client and
   // diffing the encoded messages field by field: #2 is the OUTPUT cap and #3 is
-  // the context window. This layout had those two swapped, so a caller asking
-  // for 32 output tokens put 32 into the context-window field and the request
-  // came back as an opaque "an internal error occurred" — for every turn, on
-  // every account, which is why free and paid failed identically. #6 and #11
-  // are not part of the message the service accepts.
+  // max_newlines. This layout once had those two swapped, so a caller's output
+  // cap landed in #3 and a large value in #2, and the request came back as an
+  // opaque "an internal error occurred" — for every turn, on every account,
+  // which is why free and paid failed identically. #6 and #11 are not part of
+  // the message the service accepts.
   return Buffer.concat([
     encodeVarintField(1, 1),
     encodeVarintField(2, opts.maxOutputTokens ?? DEFAULT_MAX_OUTPUT_TOKENS),
-    encodeVarintField(3, opts.maxInputTokens ?? DEFAULT_CONTEXT_WINDOW),
+    encodeVarintField(3, MAX_NEWLINES),
     enc64(5, safeTemperature(opts.temperature)),
     encodeVarintField(7, opts.topK ?? 40),
     enc64(8, opts.topP ?? 1.0),
@@ -543,7 +547,6 @@ interface BuildArgs {
   requestType?: number;
   completionOpts?: {
     maxOutputTokens?: number;
-    maxInputTokens?: number;
     temperature?: number;
     topK?: number;
     topP?: number;