@oh-my-pi/pi-catalog 18.6.0 → 18.6.3

package/src/compat/rules/classes/openai.kdl

@@ -87,3 +87,105 @@ class "openai" {
 		}
 	}
 }
+
+// GPT-5.2+ bills images in 32px patches × 1.2, sized per `detail` level:
+// https://developers.openai.com/api/docs/guides/images-vision#calculating-costs
+// Older lines carry no lineage rule and bill at their wire API's fallback
+// (providers/image-tokenization.kdl).
+class "openai" {
+	// gpt-5.6 sizing, shared by later lines. `<10` keeps separator-collapsed
+	// revisions (`openai-gpt-52-codex` parses as 52) off the 5.6+ sizing.
+	revision ">=5.6 <10" {
+		image-tokenization {
+			regime "openai-patch"
+			multiplier 1.2
+			auto "original"
+			low {
+				max-edge 512
+			}
+			high {
+				max-edge 2048
+				patch-budget 2500
+			}
+			original {
+				max-edge 65535
+			}
+		}
+	}
+	revision "=5.5" {
+		image-tokenization {
+			regime "openai-patch"
+			multiplier 1.2
+			auto "original"
+			low {
+				max-edge 512
+			}
+			high {
+				max-edge 2048
+				patch-budget 2500
+			}
+			original {
+				max-edge 6000
+				patch-budget 10000
+			}
+		}
+	}
+	// gpt-5.4: `auto` sizes like `high`, and `low` keeps a 6,144-patch budget.
+	revision "=5.4" {
+		image-tokenization {
+			regime "openai-patch"
+			multiplier 1.2
+			auto "high"
+			low {
+				max-edge 2048
+				patch-budget 6144
+			}
+			high {
+				max-edge 2048
+				patch-budget 2500
+			}
+			original {
+				max-edge 6000
+				patch-budget 10000
+			}
+		}
+	}
+	// gpt-5.2 (and the 5.3 Codex SKU) size every level the same way.
+	revision ">=5.2 <5.4" {
+		image-tokenization {
+			regime "openai-patch"
+			multiplier 1.2
+			auto "high"
+			low {
+				max-edge 2048
+				patch-budget 6144
+			}
+			high {
+				max-edge 2048
+				patch-budget 6144
+			}
+			original {
+				max-edge 2048
+				patch-budget 6144
+			}
+		}
+	}
+	// residue: astra's limits differ from the rest of its revision.
+	models "*gpt-6-astra*" {
+		image-tokenization {
+			regime "openai-patch"
+			multiplier 1.2
+			auto "original"
+			low {
+				max-edge 512
+			}
+			high {
+				max-edge 65535
+				patch-budget 2500
+			}
+			original {
+				max-edge 65535
+			}
+		}
+	}
+}

package/src/compat/rules/providers/google-antigravity.kdl

@@ -2,6 +2,9 @@
 
 provider "google-antigravity" {
 	default-model "gemini-3.1-pro"
+	// `fetchAvailableModels` is the account's served roster; bundled rows it
+	// omits (e.g. Claude 5.5 on plans without it) 404 on every request (#14328).
+	dynamic-models-authoritative #true
 	kind-apis {
 		image "google-gemini-cli"
 	}

package/src/compat/rules/providers/image-tokenization.kdl

@@ -0,0 +1,42 @@
+// Image billing for readers whose lineage has no `image-tokenization` rule in
+// classes/*.kdl, by the wire API carrying the request. priority=-1: these are
+// fallbacks, so a class-wide lineage rule (same specificity) always wins.
+
+// OpenAI-compatible wires bill like GPT-5.5: 32px patches × 1.2, `auto` sized
+// like `original`.
+on-api "openai-completions" "openai-responses" "openai-codex-responses" "azure-openai-responses" priority=-1 {
+	image-tokenization {
+		regime "openai-patch"
+		multiplier 1.2
+		auto "original"
+		low {
+			max-edge 512
+		}
+		high {
+			max-edge 2048
+			patch-budget 2500
+		}
+		original {
+			max-edge 6000
+			patch-budget 10000
+		}
+	}
+}
+
+// Google wires bill Gemini 3's fixed `media_resolution` budget per image.
+on-api "google-generative-ai" "google-gemini-cli" "google-vertex" priority=-1 {
+	image-tokenization {
+		regime "fixed"
+		tokens 1120
+	}
+}
+
+// Anthropic wires and every other adapter bill Claude's high-res tier, the
+// largest per-image bill, so an unpriced reader is never undercounted.
+on-api "anthropic-messages" "bedrock-converse-stream" "openrouter" "ollama-chat" "cursor-agent" "factory-droid-agent" "gitlab-duo-agent" "devin-agent" "apple-foundation-models" priority=-1 {
+	image-tokenization {
+		regime "anthropic-patch"
+		max-edge 2576
+		max-tokens 4784
+	}
+}

package/src/compat/rules/providers/muse-code.kdl

@@ -15,6 +15,13 @@ provider "muse-code" {
 	// api.meta.ai/v1 accepts only tool_choice "auto"; "none", "required" and named
 	// choices are rejected with 400 (verified 2026-09-10), so omit the field.
 	supports-tool-choice #false
+	// Results can be stored so a dropped stream is recovered: verified 2026-09-29
+	// against api.meta.ai/v1, a stored run keeps generating after the client
+	// socket dies and `GET /responses/{id}` returns it once finished (~60s for
+	// a high-effort turn), while `store: false` runs 404 forever. Storage is
+	// opt-in (privacy: stored runs retain prompts/outputs on Meta's side): omp
+	// setting providers.muse-code.storeResponses or PI_MUSE_STORE_RESPONSES=1.
+	store-responses #true
 	// Subscription tier bills per request; the compact edit prompt cuts the
 	// recurring tool-schema bytes. Verified 2026-09-05 against api.meta.ai/v1:
 	// `custom` tools are rejected with 400 "`custom` tools are not supported

package/src/compat/rules/providers/openai-codex.kdl

@@ -3,6 +3,7 @@
 provider "openai-codex" {
 	default-model "gpt-5.5"
 	env "OPENAI_CODEX_OAUTH_TOKEN"
+	dynamic-models-authoritative #true
 	kind-apis {
 		image "openai-codex-responses"
 	}

package/src/compat/rules/providers/snowflake.kdl

@@ -0,0 +1,135 @@
+// Documented account-billed Cortex REST API, not verified CoCo-subscription usage.
+// Model IDs and compatibility: https://docs.snowflake.com/en/user-guide/snowflake-cortex/cortex-rest-api
+// Output is capped at the Cortex REST API's documented 16,384-token maximum (rate-limit table;
+// newer models absent from that table are assumed to share it). Context windows are conservative
+// AI_COMPLETE configured ceilings, not verified REST maxima:
+// https://docs.snowflake.com/en/user-guide/snowflake-cortex/cortex-rest-api#label-cortex-rest-api-rate-limits
+// https://docs.snowflake.com/en/user-guide/snowflake-cortex/aisql-regional-availability
+// USD per million tokens estimates: Consumption Table 6(b) AI Credits x $2.00
+// (on-demand global routing); regional routing and contract discounts change billing.
+// https://www.snowflake.com/legal-files/CreditConsumptionTable.pdf
+// https://docs.snowflake.com/en/user-guide/snowflake-cortex/pricing
+provider "snowflake" {
+	default-model "claude-opus-5-5"
+	env "SNOWFLAKE_PAT"
+
+	// Credential-scoped roster: never fetch during catalog generation. The `.invalid`
+	// origin is a placeholder that snowflakeTransport replaces with the account URL.
+	seed api="anthropic-messages" base-url="https://snowflake-account.invalid/api/v2/cortex" bundle="always" precedence="seed" {
+		model "claude-opus-5-5" name="Claude Opus 5.5" {
+			reasoning #true
+			input "text" "image"
+			cost input=4 output=20 cache-write=5 cache-read=0.2
+			limits context=1000000 max-tokens=16384
+		}
+		model "claude-opus-5" name="Claude Opus 5" {
+			reasoning #true
+			input "text" "image"
+			cost input=5 output=25 cache-write=6.25 cache-read=0.5
+			limits context=1000000 max-tokens=16384
+		}
+		model "claude-opus-4-8" name="Claude Opus 4.8" {
+			reasoning #true
+			input "text" "image"
+			cost input=5 output=25 cache-write=6.25 cache-read=0.5
+			limits context=1000000 max-tokens=16384
+		}
+		model "claude-opus-4-7" name="Claude Opus 4.7" {
+			reasoning #true
+			input "text" "image"
+			cost input=5 output=25 cache-write=6.25 cache-read=0.5
+			limits context=1000000 max-tokens=16384
+		}
+		model "claude-opus-4-6" name="Claude Opus 4.6" {
+			reasoning #true
+			input "text" "image"
+			cost input=5 output=25 cache-write=6.25 cache-read=0.5
+			limits context=1000000 max-tokens=16384
+		}
+		model "claude-opus-4-5" name="Claude Opus 4.5" {
+			reasoning #true
+			input "text" "image"
+			cost input=5 output=25 cache-write=6.25 cache-read=0.5
+			limits context=200000 max-tokens=16384
+		}
+		model "claude-sonnet-5" name="Claude Sonnet 5" {
+			reasoning #true
+			input "text" "image"
+			cost input=2 output=10 cache-write=2.5 cache-read=0.2
+			limits context=1000000 max-tokens=16384
+		}
+		model "claude-sonnet-4-6" name="Claude Sonnet 4.6" {
+			reasoning #true
+			input "text" "image"
+			cost input=3 output=15 cache-write=3.75 cache-read=0.3
+			limits context=1000000 max-tokens=16384
+		}
+		model "claude-sonnet-4-5" name="Claude Sonnet 4.5" {
+			reasoning #true
+			input "text" "image"
+			cost input=3 output=15 cache-write=3.75 cache-read=0.3
+			limits context=200000 max-tokens=16384
+		}
+		model "claude-haiku-4-5" name="Claude Haiku 4.5" {
+			reasoning #true
+			input "text" "image"
+			cost input=1 output=5 cache-write=1.25 cache-read=0.1
+			limits context=200000 max-tokens=16384
+		}
+		model "openai-gpt-5.1" name="GPT-5.1" api="openai-completions" base-url="https://snowflake-account.invalid/api/v2/cortex/v1" {
+			reasoning #true
+			input "text" "image"
+			cost input=1.25 output=10 cache-write=0 cache-read=0.13
+			limits context=272000 max-tokens=16384
+		}
+		model "openai-gpt-5" name="GPT-5" api="openai-completions" base-url="https://snowflake-account.invalid/api/v2/cortex/v1" {
+			reasoning #true
+			input "text" "image"
+			cost input=1.25 output=10 cache-write=0 cache-read=0.13
+			limits context=272000 max-tokens=16384
+		}
+		model "openai-gpt-5-mini" name="GPT-5 Mini" api="openai-completions" base-url="https://snowflake-account.invalid/api/v2/cortex/v1" {
+			reasoning #true
+			input "text" "image"
+			cost input=0.25 output=2 cache-write=0 cache-read=0.03
+			limits context=272000 max-tokens=16384
+		}
+		model "openai-gpt-5-nano" name="GPT-5 Nano" api="openai-completions" base-url="https://snowflake-account.invalid/api/v2/cortex/v1" {
+			reasoning #true
+			input "text" "image"
+			cost input=0.06 output=0.44 cache-write=0 cache-read=0.01
+			limits context=272000 max-tokens=16384
+		}
+		model "openai-gpt-4.1" name="GPT-4.1" api="openai-completions" base-url="https://snowflake-account.invalid/api/v2/cortex/v1" {
+			reasoning #false
+			input "text" "image"
+			cost input=2 output=8 cache-write=0 cache-read=0.5
+			limits context=128000 max-tokens=16384
+		}
+	}
+
+	// Claude thinking modes and effort ladders come from the Anthropic lineage rules
+	// (classes/anthropic.kdl). Live Cortex probes matched them: Opus 4.7+/Sonnet 5
+	// accept xhigh and max; Opus 4.6 and Sonnet 4.6 reject xhigh.
+	// Cortex compatibility chart: ephemeral 5-minute caching and Bedrock-compatible betas only.
+	class "anthropic" {
+		supports-eager-tool-input-streaming #false
+		supports-long-cache-retention #false
+	}
+	// Cortex documents reasoning_effort none/minimal/low/medium/high. Prefixed ids
+	// (openai-gpt-*) carry no `gpt` family, so the <5.2 lineage ladder is restated
+	// here; 5.2+ rows fall through to the unscoped lineage ladders in classes/openai.kdl.
+	class "openai" {
+		revision ">=5 <5.2" {
+			thinking-mode "effort"
+			thinking-efforts "minimal" "low" "medium" "high"
+		}
+	}
+	// Cortex compatibility chart: max_tokens/store are hard errors; non-OpenAI backends accept user/assistant/system roles only.
+	on-api "openai-completions" {
+		max-tokens-field "max_completion_tokens"
+		supports-store #false
+		supports-developer-role #false
+		supports-reasoning-effort #true
+	}
+}