Never stop coding. Free MIT AI gateway: one endpoint, 375 providers (150+ free), 1,400+ models Kimi, Claude, GPT, Gemini, GLM, DeepSeek, MiniMax. Works with Claude Code, Codex, Cursor, OpenCode, Cline & Copilot. Quota-aware auto-fallback, RTK+Caveman compression saves 15-95% tokens, MCP/A2A, Desktop/PWA. Built by 900+ contributors
TypeScript a2aai-agentsai-gatewayanthropicclaudeclaude-codeclinecodexcopilotcursordeepseekfree-aigeminikimillm-gatewaymcpopenaiopenai-proxyqwentoken-saver
/**
* Build-local capability/context/override resolution snapshot (#9199).
*
* Catalog preparation bulk-loads the capability and custom-model tables once into a
* build-local view for pure in-memory resolution. This must not flip models.dev's
* module-global all-row cache, and ordinary runtime callers keep on-demand DB reads.
*
* Override maps are nested by provider then model so provider/model pairs cannot
* collide via delimiter composition.
*/
import { listModelCapabilityOverrides } from "@/lib/db/modelCapabilityOverrides";
import type { ReasoningEffortOverrideValue } from "@/shared/reasoning/reasoningEffortsOverride";
import { listModelContextOverrides } from "@/lib/db/modelContextOverrides";
import {
listModelCompatVisionOverrides,
type ModelCompatVisionOverrideMap,
} from "@/lib/db/models/compat";
import {
buildCustomModelVisionOverrides,
readCustomModelsByProvider,
type CustomModelsByProvider,
type CustomModelVisionOverrideMap,
type CustomModelVisionOverrideReadOptions,
listSyncedAvailableModelVision,
type SyncedAvailableModelVisionMap,
} from "@/lib/db/models";
import {
loadAllSyncedCapabilitiesUncached,
type CapabilitiesByProvider,
} from "@/lib/modelsDevSync";
/** Nested provider → model → numeric override map (collision-free). */
export type NestedOverrideMap = ReadonlyMap<string, ReadonlyMap<string, number>>;
export type NestedReasoningEffortsOverrideMap = ReadonlyMap<
string,
ReadonlyMap<string, readonly ReasoningEffortOverrideValue[]>
>;
export interface ModelCapabilityResolutionSnapshot {
readonly synced: CapabilitiesByProvider;
readonly maxTokenOverrides: NestedOverrideMap;
readonly maxInputTokenOverrides: NestedOverrideMap;
readonly reasoningEffortsOverrides: NestedReasoningEffortsOverrideMap;
readonly contextOverrides: NestedOverrideMap;
readonly customVisionOverrides: CustomModelVisionOverrideMap;
/**
* The `customModels` rows the vision overrides were derived from, exposed so
* listing consumers (#16929 operator endpoint overlay) reuse this one read
* instead of querying the namespace again. Optional: hand-built snapshots
* may omit it, and consumers then fall back to their own read.
*/
readonly customModelsByProvider?: CustomModelsByProvider;
readonly compatVisionOverrides: ModelCompatVisionOverrideMap;
/** #14081: positive-only vision verdicts from synced custom-node model rows. */
readonly syncedAvailableModelVision: SyncedAvailableModelVisionMap;
}
export interface ModelCapabilityResolutionSnapshotOptions {
customModelVision?: CustomModelVisionOverrideReadOptions;
}
function setNestedOverride(
map: Map<string, Map<string, number>>,
provider: string,
modelId: string,
value: number
): void {
let byModel = map.get(provider);
if (!byModel) {
byModel = new Map();
map.set(provider, byModel);
}
byModel.set(modelId, value);
}
/**
* Load all capability/custom-model tables in one uninterrupted JS turn.
* Callers must not yield between the bulk reads if they need a coherent view;
* existing catalog generation guards remain authoritative across later yields.
*/
export function createModelCapabilityResolutionSnapshot(
options: ModelCapabilityResolutionSnapshotOptions = {}
): ModelCapabilityResolutionSnapshot {
const synced = loadAllSyncedCapabilitiesUncached();
const maxTokenOverrides = new Map<string, Map<string, number>>();
const maxInputTokenOverrides = new Map<string, Map<string, number>>();
const reasoningEffortsOverrides = new Map<
string,
Map<string, readonly ReasoningEffortOverrideValue[]>
>();
for (const entry of listModelCapabilityOverrides()) {
if (entry.key === "max_output_tokens") {
setNestedOverride(maxTokenOverrides, entry.provider, entry.modelId, entry.value);
} else if (entry.key === "max_input_tokens") {
setNestedOverride(maxInputTokenOverrides, entry.provider, entry.modelId, entry.value);
} else if (entry.key === "reasoning_efforts") {
let byModel = reasoningEffortsOverrides.get(entry.provider);
if (!byModel) {
byModel = new Map();
reasoningEffortsOverrides.set(entry.provider, byModel);
}
byModel.set(entry.modelId, entry.value);
}
}
const contextOverrides = new Map<string, Map<string, number>>();
for (const entry of listModelContextOverrides()) {
setNestedOverride(contextOverrides, entry.provider, entry.modelId, entry.realContext);
}
const customModelsByProvider = readCustomModelsByProvider(options.customModelVision);
return {
synced,
maxTokenOverrides,
maxInputTokenOverrides,
reasoningEffortsOverrides,
contextOverrides,
customVisionOverrides: buildCustomModelVisionOverrides(customModelsByProvider),
customModelsByProvider,
compatVisionOverrides: listModelCompatVisionOverrides(),
syncedAvailableModelVision: listSyncedAvailableModelVision(),
};
}