mirror of
https://github.com/openclaw/openclaw.git
synced 2026-08-23 10:55:31 -06:00
refactor(providers): return prepared dynamic models directly (#126574)
This commit is contained in:
Peter Steinberger
committed by
GitHub
GitHub
parent
4324ba2f48
commit
01e8887959
@@ -272,7 +272,7 @@ listed here.
|
||||
| `resolveExternalAuthProfiles` | Overlay provider-owned external auth profiles; default `persistence` is `runtime-only` for CLI/app-owned creds | Provider reuses external auth credentials without persisting copied refresh tokens; declare `contracts.externalAuthProviders` in the manifest |
|
||||
| `shouldDeferSyntheticProfileAuth` | Lower stored synthetic profile placeholders behind env/config-backed auth | Provider stores synthetic placeholder profiles that should not win precedence |
|
||||
| `resolveDynamicModel` | Sync fallback for provider-owned model ids not in the local registry yet | Provider accepts arbitrary upstream model ids |
|
||||
| `prepareDynamicModel` | Async warm-up, then `resolveDynamicModel` runs again | Provider needs network metadata before resolving unknown ids |
|
||||
| `prepareDynamicModel` | Return an asynchronously prepared model, or warm reusable metadata before retrying `resolveDynamicModel` | Provider needs network metadata before resolving unknown ids |
|
||||
| `normalizeResolvedModel` | Final rewrite before the embedded runner uses the resolved model | Provider needs transport rewrites but still uses a core transport |
|
||||
| `normalizeToolSchemas` | Normalize tool schemas before the embedded runner sees them | Provider needs transport-family schema cleanup |
|
||||
| `inspectToolSchemas` | Surface provider-owned schema diagnostics after normalization | Provider wants keyword warnings without teaching core provider-specific rules |
|
||||
|
||||
@@ -485,8 +485,10 @@ catalog, API-key auth, and dynamic model resolution.
|
||||
});
|
||||
```
|
||||
|
||||
If resolving requires a network call, use `prepareDynamicModel` for async
|
||||
warm-up - `resolveDynamicModel` runs again after it completes.
|
||||
If resolving requires a network call, return the requested model directly
|
||||
from `prepareDynamicModel`. OpenClaw applies the same configured overrides
|
||||
and normalization as synchronous dynamic resolution. Existing hooks that
|
||||
return nothing still retry `resolveDynamicModel` after preparation.
|
||||
|
||||
</Step>
|
||||
|
||||
@@ -683,7 +685,7 @@ catalog, API-key auth, and dynamic model resolution.
|
||||
| `resolveExternalAuthProfiles` | Overlay provider-owned external auth profiles for CLI/app-managed credentials |
|
||||
| `shouldDeferSyntheticProfileAuth` | Lower synthetic stored-profile placeholders behind env/config auth |
|
||||
| `resolveDynamicModel` | Accept arbitrary upstream model IDs |
|
||||
| `prepareDynamicModel` | Async metadata fetch before resolving |
|
||||
| `prepareDynamicModel` | Return an asynchronously discovered model, or warm reusable metadata before sync resolution |
|
||||
| `normalizeResolvedModel` | Transport rewrites before the runner |
|
||||
| `normalizeToolSchemas` | Provider-owned tool-schema cleanup before registration |
|
||||
| `inspectToolSchemas` | Provider-owned tool-schema diagnostics |
|
||||
|
||||
@@ -18,6 +18,7 @@ import type { ProviderPlugin } from "openclaw/plugin-sdk/provider-model-shared";
|
||||
import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
|
||||
|
||||
const mocks = vi.hoisted(() => ({
|
||||
discoverServer: vi.fn(),
|
||||
ensureModel: vi.fn(),
|
||||
prepareServer: vi.fn(),
|
||||
inspectRuntime: vi.fn(),
|
||||
@@ -36,6 +37,11 @@ vi.mock("./src/managed-server.js", async (importOriginal) => ({
|
||||
inspectLlamaServerRuntime: mocks.inspectRuntime,
|
||||
}));
|
||||
|
||||
vi.mock("./src/external-server/discovery.js", async (importOriginal) => ({
|
||||
...(await importOriginal<typeof import("./src/external-server/discovery.js")>()),
|
||||
discoverLlamaServer: mocks.discoverServer,
|
||||
}));
|
||||
|
||||
import llamaCppPlugin from "./index.js";
|
||||
import {
|
||||
DEFAULT_LLAMA_CPP_EMBEDDING_CACHE_FILE,
|
||||
@@ -52,6 +58,7 @@ let previousPluginRegistry: ReturnType<typeof getActivePluginRegistry>;
|
||||
|
||||
beforeEach(() => {
|
||||
previousPluginRegistry = getActivePluginRegistry();
|
||||
mocks.discoverServer.mockReset();
|
||||
mocks.ensureModel.mockResolvedValue("/models/model.gguf");
|
||||
mocks.prepareServer.mockResolvedValue({});
|
||||
mocks.inspectRuntime.mockResolvedValue({
|
||||
@@ -171,6 +178,23 @@ describe("llama.cpp provider plugin", () => {
|
||||
expect(provider).not.toHaveProperty("createStreamFn");
|
||||
});
|
||||
|
||||
it("never discovers external models for a managed local service", async () => {
|
||||
const provider = registerTextProvider();
|
||||
const prepareDynamicModel = expectDefined(provider.prepareDynamicModel, "dynamic model hook");
|
||||
const { config } = configuredOptions();
|
||||
|
||||
await expect(
|
||||
prepareDynamicModel({
|
||||
config,
|
||||
provider: LLAMA_CPP_PROVIDER_ID,
|
||||
modelId: "gemma-4-e4b-it-q4_k_m",
|
||||
modelRegistry: {} as never,
|
||||
providerConfig: config.models.providers[LLAMA_CPP_PROVIDER_ID],
|
||||
}),
|
||||
).resolves.toBeUndefined();
|
||||
expect(mocks.discoverServer).not.toHaveBeenCalled();
|
||||
});
|
||||
|
||||
it("registers local embeddings through the generic provider contract", () => {
|
||||
const { config, registry } = createPluginRegistryFixture();
|
||||
registerVirtualTestPlugin({
|
||||
|
||||
@@ -3,11 +3,7 @@ import type {
|
||||
ProviderPrepareDynamicModelContext,
|
||||
} from "openclaw/plugin-sdk/plugin-entry";
|
||||
import { beforeEach, describe, expect, it, vi } from "vitest";
|
||||
import {
|
||||
discoverLlamaServerProvider,
|
||||
prepareLlamaServerDynamicModels,
|
||||
resolveLlamaServerDynamicModel,
|
||||
} from "./provider.js";
|
||||
import { discoverLlamaServerProvider, prepareLlamaServerDynamicModel } from "./provider.js";
|
||||
|
||||
const discoverMock = vi.hoisted(() => vi.fn());
|
||||
const runtimeApiKeyMock = vi.hoisted(() => vi.fn());
|
||||
@@ -64,6 +60,22 @@ function catalogContext(): ProviderCatalogContext {
|
||||
};
|
||||
}
|
||||
|
||||
function dynamicContext(
|
||||
overrides: Partial<ProviderPrepareDynamicModelContext> = {},
|
||||
): ProviderPrepareDynamicModelContext {
|
||||
return {
|
||||
config: {},
|
||||
provider: "llama-cpp",
|
||||
modelId: "org/model:Q4",
|
||||
modelRegistry: {} as never,
|
||||
providerConfig: {
|
||||
baseUrl: "http://localhost:8080/v1",
|
||||
api: "openai-completions",
|
||||
},
|
||||
...overrides,
|
||||
};
|
||||
}
|
||||
|
||||
describe("llama-server provider discovery", () => {
|
||||
beforeEach(() => {
|
||||
discoverMock.mockReset();
|
||||
@@ -131,7 +143,29 @@ describe("llama-server provider discovery", () => {
|
||||
});
|
||||
});
|
||||
|
||||
it("scopes dynamic catalogs by agent runtime and auth profile", async () => {
|
||||
it("returns only the requested discovered model directly to its preparation owner", async () => {
|
||||
discoverMock.mockResolvedValue({
|
||||
...success(),
|
||||
models: [
|
||||
model(),
|
||||
{
|
||||
...model(),
|
||||
config: { ...model().config, id: "org/requested:Q8", name: "Requested model" },
|
||||
},
|
||||
],
|
||||
});
|
||||
await expect(
|
||||
prepareLlamaServerDynamicModel(dynamicContext({ modelId: "org/requested:Q8" })),
|
||||
).resolves.toMatchObject({
|
||||
provider: "llama-cpp",
|
||||
id: "org/requested:Q8",
|
||||
name: "Requested model",
|
||||
baseUrl: "http://localhost:8080/v1",
|
||||
api: "openai-completions",
|
||||
});
|
||||
});
|
||||
|
||||
it("keeps requested models and API keys isolated by agent runtime and auth profile", async () => {
|
||||
const first = success();
|
||||
const second = {
|
||||
...success(),
|
||||
@@ -143,29 +177,20 @@ describe("llama-server provider discovery", () => {
|
||||
],
|
||||
};
|
||||
discoverMock.mockResolvedValueOnce(first).mockResolvedValueOnce(second);
|
||||
const base = {
|
||||
config: {},
|
||||
provider: "llama-cpp",
|
||||
modelId: "org/model:Q4",
|
||||
modelRegistry: {},
|
||||
providerConfig: {
|
||||
baseUrl: "http://localhost:8080/v1",
|
||||
api: "openai-completions",
|
||||
},
|
||||
};
|
||||
const firstCtx = {
|
||||
...base,
|
||||
runtimeApiKeyMock
|
||||
.mockResolvedValueOnce("first-profile-key")
|
||||
.mockResolvedValueOnce("second-profile-key");
|
||||
const firstCtx = dynamicContext({
|
||||
agentRuntimeId: "runtime-one",
|
||||
authProfileId: "profile-one",
|
||||
} as unknown as ProviderPrepareDynamicModelContext;
|
||||
const secondCtx = {
|
||||
...base,
|
||||
});
|
||||
const secondCtx = dynamicContext({
|
||||
agentRuntimeId: "runtime-two",
|
||||
authProfileId: "profile-two",
|
||||
} as unknown as ProviderPrepareDynamicModelContext;
|
||||
});
|
||||
|
||||
await prepareLlamaServerDynamicModels(firstCtx);
|
||||
await prepareLlamaServerDynamicModels(secondCtx);
|
||||
const firstModel = await prepareLlamaServerDynamicModel(firstCtx);
|
||||
const secondModel = await prepareLlamaServerDynamicModel(secondCtx);
|
||||
|
||||
expect(runtimeApiKeyMock).toHaveBeenNthCalledWith(
|
||||
1,
|
||||
@@ -175,111 +200,87 @@ describe("llama-server provider discovery", () => {
|
||||
2,
|
||||
expect.objectContaining({ profileId: "profile-two" }),
|
||||
);
|
||||
expect(resolveLlamaServerDynamicModel(firstCtx)?.name).toBe("org/model:Q4");
|
||||
expect(resolveLlamaServerDynamicModel(secondCtx)?.name).toBe("second scope");
|
||||
});
|
||||
|
||||
it("bounds dynamic model snapshots by scope", async () => {
|
||||
discoverMock.mockResolvedValue(success());
|
||||
const contexts = Array.from(
|
||||
{ length: 101 },
|
||||
(_, index) =>
|
||||
({
|
||||
config: {},
|
||||
provider: "llama-cpp",
|
||||
modelId: "org/model:Q4",
|
||||
modelRegistry: {},
|
||||
agentRuntimeId: `runtime-${index}`,
|
||||
providerConfig: {
|
||||
baseUrl: "http://localhost:8080/v1",
|
||||
api: "openai-completions",
|
||||
},
|
||||
}) as unknown as ProviderPrepareDynamicModelContext,
|
||||
expect(discoverMock).toHaveBeenNthCalledWith(
|
||||
1,
|
||||
expect.objectContaining({ apiKey: "first-profile-key" }),
|
||||
);
|
||||
|
||||
for (const ctx of contexts) {
|
||||
await prepareLlamaServerDynamicModels(ctx);
|
||||
}
|
||||
|
||||
expect(resolveLlamaServerDynamicModel(contexts[0]!)).toBeUndefined();
|
||||
expect(resolveLlamaServerDynamicModel(contexts.at(-1)!)).toMatchObject({
|
||||
id: "org/model:Q4",
|
||||
});
|
||||
expect(discoverMock).toHaveBeenNthCalledWith(
|
||||
2,
|
||||
expect.objectContaining({ apiKey: "second-profile-key" }),
|
||||
);
|
||||
expect(firstModel?.name).toBe("org/model:Q4");
|
||||
expect(secondModel?.name).toBe("second scope");
|
||||
});
|
||||
|
||||
it("keeps dynamic snapshots separate when only the endpoint changes", async () => {
|
||||
it("keeps requested models separate when only the endpoint changes", async () => {
|
||||
discoverMock.mockResolvedValueOnce(success()).mockResolvedValueOnce({
|
||||
...success(),
|
||||
models: [{ ...model(), config: { ...model().config, name: "second endpoint" } }],
|
||||
});
|
||||
const base = {
|
||||
config: {},
|
||||
provider: "llama-cpp",
|
||||
modelId: "org/model:Q4",
|
||||
modelRegistry: {},
|
||||
agentRuntimeId: "endpoint-runtime",
|
||||
authProfileId: "endpoint-profile",
|
||||
};
|
||||
const first = {
|
||||
const first = dynamicContext({
|
||||
...base,
|
||||
providerConfig: { baseUrl: "http://localhost:8080/v1", api: "openai-completions" },
|
||||
} as unknown as ProviderPrepareDynamicModelContext;
|
||||
const second = {
|
||||
});
|
||||
const second = dynamicContext({
|
||||
...base,
|
||||
providerConfig: { baseUrl: "http://localhost:8081/v1", api: "openai-completions" },
|
||||
} as unknown as ProviderPrepareDynamicModelContext;
|
||||
|
||||
await prepareLlamaServerDynamicModels(first);
|
||||
await prepareLlamaServerDynamicModels(second);
|
||||
|
||||
expect(resolveLlamaServerDynamicModel(first)?.name).toBe("org/model:Q4");
|
||||
expect(resolveLlamaServerDynamicModel(second)?.name).toBe("second endpoint");
|
||||
});
|
||||
|
||||
it("clears a scope snapshot when its refresh cannot discover the server", async () => {
|
||||
discoverMock.mockResolvedValueOnce(success()).mockResolvedValueOnce({
|
||||
kind: "unreachable",
|
||||
endpoint: { origin: "http://localhost:8080", inferenceBaseUrl: "http://localhost:8080/v1" },
|
||||
error: new Error("offline"),
|
||||
});
|
||||
const ctx = {
|
||||
config: {},
|
||||
provider: "llama-cpp",
|
||||
modelId: "org/model:Q4",
|
||||
modelRegistry: {},
|
||||
agentRuntimeId: "failed-refresh-runtime",
|
||||
providerConfig: {
|
||||
baseUrl: "http://localhost:8080/v1",
|
||||
api: "openai-completions",
|
||||
},
|
||||
} as unknown as ProviderPrepareDynamicModelContext;
|
||||
|
||||
await prepareLlamaServerDynamicModels(ctx);
|
||||
expect(resolveLlamaServerDynamicModel(ctx)).toMatchObject({ id: "org/model:Q4" });
|
||||
await prepareLlamaServerDynamicModels(ctx);
|
||||
expect(resolveLlamaServerDynamicModel(ctx)).toBeUndefined();
|
||||
});
|
||||
const firstModel = await prepareLlamaServerDynamicModel(first);
|
||||
const secondModel = await prepareLlamaServerDynamicModel(second);
|
||||
|
||||
it("refreshes and resolves dynamic model ids containing slashes", async () => {
|
||||
discoverMock.mockResolvedValue(success());
|
||||
const ctx = {
|
||||
config: {},
|
||||
provider: "llama-cpp",
|
||||
modelId: "org/model:Q4",
|
||||
modelRegistry: {},
|
||||
providerConfig: {
|
||||
baseUrl: "http://localhost:8080/v1",
|
||||
api: "openai-completions",
|
||||
},
|
||||
} as unknown as ProviderPrepareDynamicModelContext;
|
||||
|
||||
await prepareLlamaServerDynamicModels(ctx);
|
||||
|
||||
expect(resolveLlamaServerDynamicModel(ctx)).toMatchObject({
|
||||
provider: "llama-cpp",
|
||||
id: "org/model:Q4",
|
||||
expect(firstModel).toMatchObject({
|
||||
name: "org/model:Q4",
|
||||
baseUrl: "http://localhost:8080/v1",
|
||||
api: "openai-completions",
|
||||
});
|
||||
expect(secondModel).toMatchObject({
|
||||
name: "second endpoint",
|
||||
baseUrl: "http://localhost:8081/v1",
|
||||
});
|
||||
});
|
||||
|
||||
it("prefers explicit Authorization over the profile API key during model preparation", async () => {
|
||||
runtimeApiKeyMock.mockResolvedValue("profile-key");
|
||||
discoverMock.mockResolvedValue(success());
|
||||
const headers = { Authorization: "Bearer endpoint-key" };
|
||||
|
||||
await prepareLlamaServerDynamicModel(
|
||||
dynamicContext({
|
||||
providerConfig: { baseUrl: "http://localhost:8080/v1", headers },
|
||||
}),
|
||||
);
|
||||
|
||||
expect(discoverMock).toHaveBeenCalledWith(
|
||||
expect.objectContaining({ apiKey: undefined, headers, cacheTtlMs: 0 }),
|
||||
);
|
||||
});
|
||||
|
||||
it.each([
|
||||
{
|
||||
label: "the server is unavailable",
|
||||
discovery: {
|
||||
kind: "unreachable" as const,
|
||||
endpoint: { origin: "http://localhost:8080", inferenceBaseUrl: "http://localhost:8080/v1" },
|
||||
error: new Error("offline"),
|
||||
},
|
||||
},
|
||||
{
|
||||
label: "the requested model disappears",
|
||||
discovery: { ...success(), models: [] },
|
||||
},
|
||||
])("returns no stale model when $label", async ({ discovery }) => {
|
||||
discoverMock.mockResolvedValueOnce(success()).mockResolvedValueOnce(discovery);
|
||||
const ctx = dynamicContext({
|
||||
agentRuntimeId: "failed-refresh-runtime",
|
||||
});
|
||||
|
||||
await expect(prepareLlamaServerDynamicModel(ctx)).resolves.toMatchObject({
|
||||
id: "org/model:Q4",
|
||||
});
|
||||
await expect(prepareLlamaServerDynamicModel(ctx)).resolves.toBeUndefined();
|
||||
});
|
||||
});
|
||||
|
||||
@@ -1,8 +1,6 @@
|
||||
import { pruneMapToMaxSize } from "openclaw/plugin-sdk/collection-runtime";
|
||||
import type {
|
||||
ProviderCatalogContext,
|
||||
ProviderPrepareDynamicModelContext,
|
||||
ProviderResolveDynamicModelContext,
|
||||
ProviderRuntimeModel,
|
||||
} from "openclaw/plugin-sdk/plugin-entry";
|
||||
import type { ModelProviderConfig } from "openclaw/plugin-sdk/provider-model-shared";
|
||||
@@ -14,47 +12,7 @@ import {
|
||||
} from "./auth.js";
|
||||
import { discoverLlamaServer } from "./discovery.js";
|
||||
import { resolveLlamaServerEndpoint } from "./endpoint.js";
|
||||
import { buildLlamaServerProviderConfig, type LlamaServerDiscoveredModel } from "./models.js";
|
||||
|
||||
const dynamicModels = new Map<string, ProviderRuntimeModel[]>();
|
||||
const LLAMA_SERVER_DYNAMIC_MODEL_MAX_SCOPES = 100;
|
||||
|
||||
function cacheDynamicModels(key: string, models: ProviderRuntimeModel[]): void {
|
||||
dynamicModels.delete(key);
|
||||
dynamicModels.set(key, models);
|
||||
pruneMapToMaxSize(dynamicModels, LLAMA_SERVER_DYNAMIC_MODEL_MAX_SCOPES);
|
||||
}
|
||||
|
||||
function dynamicModelScopeKey(
|
||||
ctx: Pick<
|
||||
ProviderResolveDynamicModelContext,
|
||||
"agentRuntimeId" | "agentDir" | "authProfileId" | "providerConfig"
|
||||
>,
|
||||
): string {
|
||||
return [
|
||||
ctx.agentRuntimeId ?? ctx.agentDir ?? "",
|
||||
ctx.authProfileId ?? "",
|
||||
ctx.providerConfig?.baseUrl ?? "",
|
||||
].join("\u0000");
|
||||
}
|
||||
|
||||
function toRuntimeModel(
|
||||
model: LlamaServerDiscoveredModel,
|
||||
providerConfig: {
|
||||
baseUrl?: string;
|
||||
api?: ProviderRuntimeModel["api"];
|
||||
},
|
||||
): ProviderRuntimeModel {
|
||||
return {
|
||||
...model.config,
|
||||
provider: LLAMA_CPP_PROVIDER_ID,
|
||||
api: providerConfig.api ?? "openai-completions",
|
||||
baseUrl: resolveLlamaServerEndpoint(providerConfig.baseUrl).inferenceBaseUrl,
|
||||
input: model.config.input.filter(
|
||||
(entry): entry is "text" | "image" => entry === "text" || entry === "image",
|
||||
),
|
||||
};
|
||||
}
|
||||
import { buildLlamaServerProviderConfig } from "./models.js";
|
||||
|
||||
/** Discovers external llama-server models for provider runtime resolution. */
|
||||
export async function discoverLlamaServerProvider(
|
||||
@@ -96,9 +54,9 @@ export async function discoverLlamaServerProvider(
|
||||
};
|
||||
}
|
||||
|
||||
export async function prepareLlamaServerDynamicModels(
|
||||
export async function prepareLlamaServerDynamicModel(
|
||||
ctx: ProviderPrepareDynamicModelContext,
|
||||
): Promise<void> {
|
||||
): Promise<ProviderRuntimeModel | undefined> {
|
||||
const apiKey = await resolveLlamaServerRuntimeApiKey({
|
||||
config: ctx.config,
|
||||
agentDir: ctx.agentDir,
|
||||
@@ -115,19 +73,20 @@ export async function prepareLlamaServerDynamicModels(
|
||||
headers,
|
||||
cacheTtlMs: 0,
|
||||
});
|
||||
const key = dynamicModelScopeKey(ctx);
|
||||
cacheDynamicModels(
|
||||
key,
|
||||
const model =
|
||||
discovery.kind === "success"
|
||||
? discovery.models.map((model) => toRuntimeModel(model, ctx.providerConfig ?? {}))
|
||||
: [],
|
||||
);
|
||||
}
|
||||
|
||||
export function resolveLlamaServerDynamicModel(
|
||||
params: ProviderResolveDynamicModelContext,
|
||||
): ProviderRuntimeModel | undefined {
|
||||
return dynamicModels
|
||||
.get(dynamicModelScopeKey(params))
|
||||
?.find((model) => model.id === params.modelId);
|
||||
? discovery.models.find((entry) => entry.config.id === ctx.modelId)
|
||||
: undefined;
|
||||
if (!model) {
|
||||
return undefined;
|
||||
}
|
||||
return {
|
||||
...model.config,
|
||||
provider: LLAMA_CPP_PROVIDER_ID,
|
||||
api: ctx.providerConfig?.api ?? "openai-completions",
|
||||
baseUrl: resolveLlamaServerEndpoint(ctx.providerConfig?.baseUrl).inferenceBaseUrl,
|
||||
input: model.config.input.filter(
|
||||
(entry): entry is "text" | "image" => entry === "text" || entry === "image",
|
||||
),
|
||||
};
|
||||
}
|
||||
|
||||
@@ -21,8 +21,7 @@ import {
|
||||
import { normalizeLlamaServerProviderConfig } from "./external-server/endpoint.js";
|
||||
import {
|
||||
discoverLlamaServerProvider,
|
||||
prepareLlamaServerDynamicModels,
|
||||
resolveLlamaServerDynamicModel,
|
||||
prepareLlamaServerDynamicModel,
|
||||
} from "./external-server/provider.js";
|
||||
import {
|
||||
configureLlamaServerNonInteractive,
|
||||
@@ -119,15 +118,10 @@ export function registerLlamaCppProvider(api: OpenClawPluginApi): void {
|
||||
providerConfig.localService
|
||||
? providerConfig
|
||||
: normalizeLlamaServerProviderConfig(providerConfig),
|
||||
prepareDynamicModel: async (ctx) => {
|
||||
if (!ctx.config?.models?.providers?.[LLAMA_CPP_PROVIDER_ID]?.localService) {
|
||||
await prepareLlamaServerDynamicModels(ctx);
|
||||
}
|
||||
},
|
||||
resolveDynamicModel: (ctx) =>
|
||||
prepareDynamicModel: async (ctx) =>
|
||||
ctx.config?.models?.providers?.[LLAMA_CPP_PROVIDER_ID]?.localService
|
||||
? undefined
|
||||
: resolveLlamaServerDynamicModel(ctx),
|
||||
: await prepareLlamaServerDynamicModel(ctx),
|
||||
wrapStreamFn: (ctx) => {
|
||||
const providerConfig = ctx.config?.models?.providers?.[LLAMA_CPP_PROVIDER_ID];
|
||||
if (!providerConfig?.localService) {
|
||||
|
||||
@@ -25,7 +25,7 @@ export {
|
||||
normalizeLmstudioConfiguredCatalogEntry,
|
||||
normalizeLmstudioProviderConfig,
|
||||
prepareAppGuidedLmstudioSetup,
|
||||
prepareLmstudioDynamicModels,
|
||||
prepareLmstudioDynamicModel,
|
||||
promptAndConfigureLmstudioInteractive,
|
||||
resolveLmstudioConfiguredApiKey,
|
||||
resolveLmstudioInferenceBase,
|
||||
|
||||
@@ -1,17 +1,29 @@
|
||||
// Lmstudio tests cover index plugin behavior.
|
||||
import type { OpenClawConfig, ProviderAuthMethod } from "openclaw/plugin-sdk/plugin-entry";
|
||||
import { createDeferred } from "openclaw/plugin-sdk/extension-shared";
|
||||
import type {
|
||||
OpenClawConfig,
|
||||
ProviderAuthMethod,
|
||||
ProviderPrepareDynamicModelContext,
|
||||
} from "openclaw/plugin-sdk/plugin-entry";
|
||||
import { capturePluginRegistration } from "openclaw/plugin-sdk/plugin-test-runtime";
|
||||
import { CUSTOM_LOCAL_AUTH_MARKER } from "openclaw/plugin-sdk/provider-auth";
|
||||
import type { ModelProviderConfig } from "openclaw/plugin-sdk/provider-model-shared";
|
||||
import type {
|
||||
ModelDefinitionConfig,
|
||||
ModelProviderConfig,
|
||||
} from "openclaw/plugin-sdk/provider-model-shared";
|
||||
import { beforeEach, describe, expect, it, vi } from "vitest";
|
||||
import plugin from "./index.js";
|
||||
import { LMSTUDIO_LOCAL_API_KEY_PLACEHOLDER } from "./src/defaults.js";
|
||||
|
||||
const fetchLmstudioModelsMock = vi.hoisted(() => vi.fn());
|
||||
const discoverLmstudioModelsMock = vi.hoisted(() =>
|
||||
vi.fn<typeof import("./src/models.fetch.js").discoverLmstudioModels>(),
|
||||
);
|
||||
|
||||
vi.mock("./src/models.fetch.js", async (importOriginal) => ({
|
||||
...(await importOriginal<typeof import("./src/models.fetch.js")>()),
|
||||
fetchLmstudioModels: fetchLmstudioModelsMock,
|
||||
discoverLmstudioModels: discoverLmstudioModelsMock,
|
||||
}));
|
||||
|
||||
function registerProvider() {
|
||||
@@ -83,9 +95,42 @@ function createRemoteProviderConfig(overrides?: Partial<ModelProviderConfig>): M
|
||||
};
|
||||
}
|
||||
|
||||
function createDynamicModelContext(
|
||||
profile: "first" | "second",
|
||||
): ProviderPrepareDynamicModelContext {
|
||||
return {
|
||||
provider: "lmstudio",
|
||||
modelId: "shared-model",
|
||||
modelRegistry: {
|
||||
getAll: () => [],
|
||||
getAvailable: () => [],
|
||||
find: () => undefined,
|
||||
hasConfiguredAuth: () => false,
|
||||
},
|
||||
authProfileId: `lmstudio:${profile}`,
|
||||
providerConfig: {
|
||||
baseUrl: "http://lmstudio.internal:1234/v1",
|
||||
headers: { Authorization: `Bearer ${profile}-profile` },
|
||||
},
|
||||
};
|
||||
}
|
||||
|
||||
function createDiscoveredModel(name: string, contextWindow: number): ModelDefinitionConfig {
|
||||
return {
|
||||
id: "shared-model",
|
||||
name,
|
||||
reasoning: false,
|
||||
input: ["text"],
|
||||
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
|
||||
contextWindow,
|
||||
maxTokens: 8192,
|
||||
};
|
||||
}
|
||||
|
||||
describe("lmstudio plugin", () => {
|
||||
beforeEach(() => {
|
||||
fetchLmstudioModelsMock.mockReset();
|
||||
discoverLmstudioModelsMock.mockReset();
|
||||
});
|
||||
|
||||
it("registers llama.cpp GBNF tool-schema projection", () => {
|
||||
@@ -95,6 +140,71 @@ describe("lmstudio plugin", () => {
|
||||
});
|
||||
});
|
||||
|
||||
it("keeps concurrent model preparations isolated when shared-endpoint profiles finish in reverse order", async () => {
|
||||
const provider = registerProvider();
|
||||
const prepareDynamicModel = provider.prepareDynamicModel;
|
||||
if (!prepareDynamicModel) {
|
||||
throw new Error("expected the LM Studio provider to prepare dynamic models");
|
||||
}
|
||||
const firstDiscovery = createDeferred<ModelDefinitionConfig[]>();
|
||||
const secondDiscovery = createDeferred<ModelDefinitionConfig[]>();
|
||||
discoverLmstudioModelsMock.mockImplementation(({ headers }) =>
|
||||
headers?.Authorization === "Bearer first-profile"
|
||||
? firstDiscovery.promise
|
||||
: secondDiscovery.promise,
|
||||
);
|
||||
|
||||
const firstPreparation = prepareDynamicModel(createDynamicModelContext("first"));
|
||||
const secondPreparation = prepareDynamicModel(createDynamicModelContext("second"));
|
||||
await vi.waitFor(() => expect(discoverLmstudioModelsMock).toHaveBeenCalledTimes(2));
|
||||
|
||||
secondDiscovery.resolve([createDiscoveredModel("Second profile model", 65_536)]);
|
||||
const secondPrepared = await secondPreparation;
|
||||
firstDiscovery.resolve([createDiscoveredModel("First profile model", 32_768)]);
|
||||
const firstPrepared = await firstPreparation;
|
||||
|
||||
expect(secondPrepared).toMatchObject({
|
||||
id: "shared-model",
|
||||
name: "Second profile model",
|
||||
contextWindow: 65_536,
|
||||
});
|
||||
expect(firstPrepared).toMatchObject({
|
||||
id: "shared-model",
|
||||
name: "First profile model",
|
||||
contextWindow: 32_768,
|
||||
});
|
||||
});
|
||||
|
||||
it("returns only the requested discovered model without retaining stale endpoint results", async () => {
|
||||
const provider = registerProvider();
|
||||
const prepareDynamicModel = provider.prepareDynamicModel;
|
||||
if (!prepareDynamicModel) {
|
||||
throw new Error("expected the LM Studio provider to prepare dynamic models");
|
||||
}
|
||||
const otherModel = { ...createDiscoveredModel("Other model", 16_384), id: "other-model" };
|
||||
const requestedModel = createDiscoveredModel("Requested model", 32_768);
|
||||
discoverLmstudioModelsMock
|
||||
.mockResolvedValueOnce([otherModel, requestedModel])
|
||||
.mockResolvedValueOnce([otherModel]);
|
||||
|
||||
const context = createDynamicModelContext("first");
|
||||
await expect(prepareDynamicModel(context)).resolves.toMatchObject({
|
||||
id: "shared-model",
|
||||
name: "Requested model",
|
||||
provider: "lmstudio",
|
||||
api: "openai-completions",
|
||||
baseUrl: "http://lmstudio.internal:1234/v1",
|
||||
contextWindow: 32_768,
|
||||
});
|
||||
await expect(prepareDynamicModel(context)).resolves.toBeUndefined();
|
||||
expect(discoverLmstudioModelsMock).toHaveBeenCalledWith({
|
||||
baseUrl: "http://lmstudio.internal:1234/v1",
|
||||
apiKey: "",
|
||||
headers: { Authorization: "Bearer first-profile" },
|
||||
quiet: true,
|
||||
});
|
||||
});
|
||||
|
||||
it("preflights the requested LM Studio model before destructive non-interactive reset", async () => {
|
||||
fetchLmstudioModelsMock.mockResolvedValue({
|
||||
reachable: true,
|
||||
|
||||
@@ -8,7 +8,6 @@ import {
|
||||
type ProviderAuthMethod,
|
||||
type ProviderAuthMethodNonInteractiveContext,
|
||||
type ProviderAuthResult,
|
||||
type ProviderRuntimeModel,
|
||||
} from "openclaw/plugin-sdk/plugin-entry";
|
||||
import {
|
||||
CUSTOM_LOCAL_AUTH_MARKER,
|
||||
@@ -33,8 +32,6 @@ import { shouldUseLmstudioSyntheticAuth } from "./src/provider-auth.js";
|
||||
import { wrapLmstudioInferencePreload } from "./src/stream.js";
|
||||
|
||||
const PROVIDER_ID = "lmstudio";
|
||||
// Intentional: dynamic models are cached per LM Studio endpoint (`baseUrl`) only.
|
||||
const cachedDynamicModels = new Map<string, ProviderRuntimeModel[]>();
|
||||
|
||||
type LmstudioNonInteractiveValidationContext = Parameters<
|
||||
NonNullable<ProviderAuthMethod["validateNonInteractive"]>
|
||||
@@ -233,15 +230,8 @@ export default definePluginEntry({
|
||||
normalizeConfig: ({ providerConfig }) => normalizeLmstudioProviderConfig(providerConfig),
|
||||
prepareDynamicModel: async (ctx) => {
|
||||
const providerSetup = await loadProviderSetup();
|
||||
cachedDynamicModels.set(
|
||||
ctx.providerConfig?.baseUrl ?? "",
|
||||
await providerSetup.prepareLmstudioDynamicModels(ctx),
|
||||
);
|
||||
return await providerSetup.prepareLmstudioDynamicModel(ctx);
|
||||
},
|
||||
resolveDynamicModel: (ctx) =>
|
||||
cachedDynamicModels
|
||||
.get(ctx.providerConfig?.baseUrl ?? "")
|
||||
?.find((model) => model.id === ctx.modelId),
|
||||
augmentModelCatalog: (ctx) => resolveLmstudioAugmentedCatalogEntries(ctx.config),
|
||||
wrapStreamFn: wrapLmstudioInferencePreload,
|
||||
...buildProviderToolCompatFamilyHooks("llamacpp-gbnf"),
|
||||
|
||||
@@ -40,6 +40,6 @@ export {
|
||||
detectAppGuidedLmstudioAvailability,
|
||||
discoverLmstudioProvider,
|
||||
prepareAppGuidedLmstudioSetup,
|
||||
prepareLmstudioDynamicModels,
|
||||
prepareLmstudioDynamicModel,
|
||||
promptAndConfigureLmstudioInteractive,
|
||||
} from "./setup.js";
|
||||
|
||||
@@ -1046,9 +1046,9 @@ export async function discoverLmstudioProvider(ctx: ProviderCatalogContext): Pro
|
||||
};
|
||||
}
|
||||
|
||||
export async function prepareLmstudioDynamicModels(
|
||||
export async function prepareLmstudioDynamicModel(
|
||||
ctx: ProviderPrepareDynamicModelContext,
|
||||
): Promise<ProviderRuntimeModel[]> {
|
||||
): Promise<ProviderRuntimeModel | undefined> {
|
||||
const baseUrl = resolveLmstudioInferenceBase(ctx.providerConfig?.baseUrl);
|
||||
const { apiKey, headers } = await resolveLmstudioRequestContext({
|
||||
config: ctx.config,
|
||||
@@ -1062,15 +1062,18 @@ export async function prepareLmstudioDynamicModels(
|
||||
headers,
|
||||
quiet: true,
|
||||
});
|
||||
return discoveredModels.map((model) =>
|
||||
Object.assign({}, model, {
|
||||
provider: PROVIDER_ID,
|
||||
api: ctx.providerConfig?.api ?? `openai-completions`,
|
||||
baseUrl,
|
||||
input: model.input.filter(
|
||||
(entry): entry is "text" | "image" => entry === "text" || entry === "image",
|
||||
),
|
||||
}),
|
||||
);
|
||||
const model = discoveredModels.find((candidate) => candidate.id === ctx.modelId);
|
||||
if (!model) {
|
||||
return undefined;
|
||||
}
|
||||
return {
|
||||
...model,
|
||||
provider: PROVIDER_ID,
|
||||
api: ctx.providerConfig?.api ?? "openai-completions",
|
||||
baseUrl,
|
||||
input: model.input.filter(
|
||||
(entry): entry is "text" | "image" => entry === "text" || entry === "image",
|
||||
),
|
||||
};
|
||||
}
|
||||
/* oxlint-disable max-lines -- TODO: split this grandfathered oversized file. */
|
||||
|
||||
@@ -1084,9 +1084,7 @@ describe("ollama plugin", () => {
|
||||
const context = createDynamicModelContext("llama3.2:latest");
|
||||
|
||||
try {
|
||||
await provider.prepareDynamicModel?.(context as never);
|
||||
|
||||
const resolved = provider.resolveDynamicModel?.(context as never);
|
||||
const resolved = await provider.prepareDynamicModel?.(context as never);
|
||||
expect(resolved?.provider).toBe("ollama");
|
||||
expect(resolved?.id).toBe("llama3.2:latest");
|
||||
expect(resolved?.api).toBe("ollama");
|
||||
@@ -1134,9 +1132,7 @@ describe("ollama plugin", () => {
|
||||
};
|
||||
const context = createDynamicModelContext("qwen3-coder:cloud", config);
|
||||
|
||||
await provider.prepareDynamicModel?.(context as never);
|
||||
|
||||
const resolved = provider.resolveDynamicModel?.(context as never);
|
||||
const resolved = await provider.prepareDynamicModel?.(context as never);
|
||||
expect(resolved?.provider).toBe("ollama");
|
||||
expect(resolved?.id).toBe("qwen3-coder:cloud");
|
||||
expect(resolved?.api).toBe("openai-completions");
|
||||
@@ -1172,7 +1168,7 @@ describe("ollama plugin", () => {
|
||||
mockDiscoveredOllamaProvider([], { baseUrl, once: true });
|
||||
const context = createDynamicModelContext("private-dynamic-model", config);
|
||||
|
||||
await provider.prepareDynamicModel?.(context as never);
|
||||
const resolved = await provider.prepareDynamicModel?.(context as never);
|
||||
|
||||
expect(buildOllamaProviderMock).toHaveBeenCalledWith(baseUrl, {
|
||||
quiet: true,
|
||||
@@ -1181,10 +1177,10 @@ describe("ollama plugin", () => {
|
||||
expect(queryOllamaModelShowInfoMock).toHaveBeenCalledWith(baseUrl, "private-dynamic-model", {
|
||||
apiKey: "dynamic-discovery-access",
|
||||
});
|
||||
expect(provider.resolveDynamicModel?.(context as never)?.id).toBe("private-dynamic-model");
|
||||
expect(resolved?.id).toBe("private-dynamic-model");
|
||||
});
|
||||
|
||||
it("scopes dynamic Ollama model caches to the effective credential", async () => {
|
||||
it("returns the exact prepared Ollama model for concurrent credential profiles", async () => {
|
||||
const provider = registerProvider();
|
||||
const baseUrl = "https://shared-dynamic-ollama.example.com";
|
||||
const modelId = "tenant-dynamic-model";
|
||||
@@ -1197,25 +1193,34 @@ describe("ollama plugin", () => {
|
||||
});
|
||||
const discoveredFor = (name: string) => ({
|
||||
baseUrl,
|
||||
api: "ollama",
|
||||
api: "ollama" as const,
|
||||
models: [{ id: modelId, name, contextWindow: 8192, maxTokens: 2048 }],
|
||||
});
|
||||
buildOllamaProviderMock
|
||||
.mockResolvedValueOnce(discoveredFor("First tenant model"))
|
||||
.mockResolvedValueOnce(discoveredFor("Second tenant model"));
|
||||
const completeDiscovery: Array<(result: ReturnType<typeof discoveredFor>) => void> = [];
|
||||
buildOllamaProviderMock.mockImplementation(
|
||||
() =>
|
||||
new Promise<ReturnType<typeof discoveredFor>>((resolve) => {
|
||||
completeDiscovery.push(resolve);
|
||||
}),
|
||||
);
|
||||
|
||||
for (const config of [configFor("first-tenant-access"), configFor("second-tenant-access")]) {
|
||||
await provider.prepareDynamicModel?.(createDynamicModelContext(modelId, config) as never);
|
||||
}
|
||||
const prepareFor = (apiKey: string, authProfileId: string) =>
|
||||
provider.prepareDynamicModel?.({
|
||||
...createDynamicModelContext(modelId, configFor(apiKey)),
|
||||
authProfileId,
|
||||
} as never);
|
||||
const firstPrepared = prepareFor("first-tenant-access", "ollama:first");
|
||||
const secondPrepared = prepareFor("second-tenant-access", "ollama:second");
|
||||
|
||||
const resolveFor = (apiKey: string) =>
|
||||
provider.resolveDynamicModel?.(
|
||||
createDynamicModelContext(modelId, configFor(apiKey)) as never,
|
||||
);
|
||||
await vi.waitFor(() => expect(buildOllamaProviderMock).toHaveBeenCalledTimes(2));
|
||||
completeDiscovery[1]?.(discoveredFor("Second tenant model"));
|
||||
await expect(secondPrepared).resolves.toMatchObject({
|
||||
id: modelId,
|
||||
name: "Second tenant model",
|
||||
});
|
||||
completeDiscovery[0]?.(discoveredFor("First tenant model"));
|
||||
await expect(firstPrepared).resolves.toMatchObject({ id: modelId, name: "First tenant model" });
|
||||
|
||||
expect(resolveFor("first-tenant-access")?.name).toBe("First tenant model");
|
||||
expect(resolveFor("second-tenant-access")?.name).toBe("Second tenant model");
|
||||
expect(resolveFor("unprepared-tenant-access")).toBeUndefined();
|
||||
expect(buildOllamaProviderMock).toHaveBeenNthCalledWith(1, baseUrl, {
|
||||
quiet: true,
|
||||
apiKey: "first-tenant-access",
|
||||
@@ -1250,8 +1255,9 @@ describe("ollama plugin", () => {
|
||||
const context = createDynamicModelContext("secretref-dynamic-model", config);
|
||||
|
||||
try {
|
||||
await provider.prepareDynamicModel?.(context as never);
|
||||
const resolved = await provider.prepareDynamicModel?.(context as never);
|
||||
|
||||
expect(resolved?.id).toBe("secretref-dynamic-model");
|
||||
expect(buildOllamaProviderMock).toHaveBeenCalledWith(baseUrl, {
|
||||
quiet: true,
|
||||
apiKey: secretValue,
|
||||
@@ -1290,7 +1296,7 @@ describe("ollama plugin", () => {
|
||||
});
|
||||
|
||||
try {
|
||||
await provider.prepareDynamicModel?.(context as never);
|
||||
await expect(provider.prepareDynamicModel?.(context as never)).resolves.toBeUndefined();
|
||||
|
||||
expect(buildOllamaProviderMock).not.toHaveBeenCalled();
|
||||
expect(queryOllamaModelShowInfoMock).not.toHaveBeenCalled();
|
||||
@@ -1301,7 +1307,7 @@ describe("ollama plugin", () => {
|
||||
}
|
||||
});
|
||||
|
||||
it("invalidates managed dynamic model caches when their SecretRef stops resolving", async () => {
|
||||
it("keeps rotated managed SecretRefs request-owned and fails closed when unavailable", async () => {
|
||||
const provider = registerProvider();
|
||||
const baseUrl = "https://managed-dynamic-ollama.example.com";
|
||||
const modelId = "managed-private-model";
|
||||
@@ -1319,20 +1325,32 @@ describe("ollama plugin", () => {
|
||||
};
|
||||
resolveConfiguredSecretInputStringMock
|
||||
.mockResolvedValueOnce({ value: "managed-dynamic-access" })
|
||||
.mockResolvedValueOnce({ value: "rotated-managed-access" })
|
||||
.mockResolvedValueOnce({ unresolvedRefReason: "managed credential is unavailable" });
|
||||
mockDiscoveredOllamaProvider(
|
||||
[{ id: modelId, name: "Managed private model", contextWindow: 8192 }],
|
||||
{ baseUrl, once: true },
|
||||
);
|
||||
mockDiscoveredOllamaProvider(
|
||||
[{ id: modelId, name: "Rotated managed model", contextWindow: 8192 }],
|
||||
{ baseUrl, once: true },
|
||||
);
|
||||
const context = createDynamicModelContext(modelId, config);
|
||||
|
||||
await provider.prepareDynamicModel?.(context as never);
|
||||
expect(provider.resolveDynamicModel?.(context as never)?.id).toBe(modelId);
|
||||
|
||||
await provider.prepareDynamicModel?.(context as never);
|
||||
|
||||
expect(provider.resolveDynamicModel?.(context as never)).toBeUndefined();
|
||||
expect(buildOllamaProviderMock).toHaveBeenCalledOnce();
|
||||
await expect(provider.prepareDynamicModel?.(context as never)).resolves.toMatchObject({
|
||||
id: modelId,
|
||||
name: "Managed private model",
|
||||
});
|
||||
await expect(provider.prepareDynamicModel?.(context as never)).resolves.toMatchObject({
|
||||
id: modelId,
|
||||
name: "Rotated managed model",
|
||||
});
|
||||
await expect(provider.prepareDynamicModel?.(context as never)).resolves.toBeUndefined();
|
||||
expect(buildOllamaProviderMock).toHaveBeenNthCalledWith(2, baseUrl, {
|
||||
quiet: true,
|
||||
apiKey: "rotated-managed-access",
|
||||
});
|
||||
expect(buildOllamaProviderMock).toHaveBeenCalledTimes(2);
|
||||
});
|
||||
|
||||
it("isolates identically named managed SecretRefs by their resolved configuration", async () => {
|
||||
@@ -1371,15 +1389,11 @@ describe("ollama plugin", () => {
|
||||
);
|
||||
const contextFor = (config: typeof firstConfig) => createDynamicModelContext(modelId, config);
|
||||
|
||||
await provider.prepareDynamicModel?.(contextFor(firstConfig) as never);
|
||||
await provider.prepareDynamicModel?.(contextFor(secondConfig) as never);
|
||||
const firstModel = await provider.prepareDynamicModel?.(contextFor(firstConfig) as never);
|
||||
const secondModel = await provider.prepareDynamicModel?.(contextFor(secondConfig) as never);
|
||||
|
||||
expect(provider.resolveDynamicModel?.(contextFor(firstConfig) as never)?.name).toBe(
|
||||
"First managed tenant model",
|
||||
);
|
||||
expect(provider.resolveDynamicModel?.(contextFor(secondConfig) as never)?.name).toBe(
|
||||
"Second managed tenant model",
|
||||
);
|
||||
expect(firstModel?.name).toBe("First managed tenant model");
|
||||
expect(secondModel?.name).toBe("Second managed tenant model");
|
||||
expect(buildOllamaProviderMock).toHaveBeenNthCalledWith(1, baseUrl, {
|
||||
quiet: true,
|
||||
apiKey: "first-managed-tenant-access",
|
||||
@@ -1412,13 +1426,12 @@ describe("ollama plugin", () => {
|
||||
const context = createDynamicModelContext("deepseek-v4-pro:cloud");
|
||||
|
||||
try {
|
||||
await provider.prepareDynamicModel?.(context as never);
|
||||
const resolved = await provider.prepareDynamicModel?.(context as never);
|
||||
|
||||
expect(queryOllamaModelShowInfoMock).toHaveBeenCalledWith(
|
||||
"http://127.0.0.1:11434",
|
||||
"deepseek-v4-pro:cloud",
|
||||
);
|
||||
const resolved = provider.resolveDynamicModel?.(context as never);
|
||||
expect(resolved?.provider).toBe("ollama");
|
||||
expect(resolved?.id).toBe("deepseek-v4-pro:cloud");
|
||||
expect(resolved?.api).toBe("ollama");
|
||||
@@ -1971,9 +1984,7 @@ describe("ollama plugin", () => {
|
||||
const context = createDynamicModelContext("depseek-v4-pro:cloud");
|
||||
|
||||
try {
|
||||
await provider.prepareDynamicModel?.(context as never);
|
||||
|
||||
expect(provider.resolveDynamicModel?.(context as never)).toBeUndefined();
|
||||
await expect(provider.prepareDynamicModel?.(context as never)).resolves.toBeUndefined();
|
||||
} finally {
|
||||
if (previous === undefined) {
|
||||
delete process.env.OLLAMA_API_KEY;
|
||||
|
||||
+13
-117
@@ -1,5 +1,4 @@
|
||||
// Ollama plugin entrypoint registers its OpenClaw integration.
|
||||
import { createHash } from "node:crypto";
|
||||
import { collectConfiguredModelRefValues } from "@openclaw/model-catalog-core/configured-model-refs";
|
||||
import type { OpenClawConfig } from "openclaw/plugin-sdk/config-contracts";
|
||||
import { createLazyRuntimeModule } from "openclaw/plugin-sdk/lazy-runtime";
|
||||
@@ -161,8 +160,6 @@ function classifyOllamaFailoverReason(errorMessage: string): "server_error" | un
|
||||
return errorMessage.trim() === OLLAMA_INCOMPLETE_STREAM_ERROR ? "server_error" : undefined;
|
||||
}
|
||||
|
||||
const dynamicModelCache = new Map<string, ProviderRuntimeModel[]>();
|
||||
const dynamicManagedCredentialFingerprints = new WeakMap<OpenClawConfig, Map<string, string>>();
|
||||
const OLLAMA_CLOUD_DEFAULT_MODEL_REF = `${OLLAMA_CLOUD_PROVIDER_ID}/${OLLAMA_CLOUD_DEFAULT_MODELS[0].id}`;
|
||||
const OLLAMA_CONFIGURED_SHOW_CONCURRENCY = 4;
|
||||
const OLLAMA_CONFIGURED_SHOW_MAX_MODELS = 8;
|
||||
@@ -426,44 +423,6 @@ async function discoverAppGuidedOllamaModel(
|
||||
};
|
||||
}
|
||||
|
||||
function buildDynamicManagedSecretScope(
|
||||
provider: string,
|
||||
baseUrl: string | undefined,
|
||||
configuredApiKey: unknown,
|
||||
): string | undefined {
|
||||
const secretRef = coerceSecretRef(configuredApiKey);
|
||||
if (!secretRef || secretRef.source === "env") {
|
||||
return undefined;
|
||||
}
|
||||
return `${provider}\0${resolveOllamaApiBase(baseUrl)}\0${secretRef.source}\0${secretRef.provider}\0${secretRef.id}`;
|
||||
}
|
||||
|
||||
function buildDynamicCacheKey(
|
||||
provider: string,
|
||||
baseUrl: string | undefined,
|
||||
configuredApiKey: unknown,
|
||||
config?: OpenClawConfig,
|
||||
): string {
|
||||
const secretRef = coerceSecretRef(configuredApiKey);
|
||||
const managedSecretScope = buildDynamicManagedSecretScope(provider, baseUrl, configuredApiKey);
|
||||
const apiKey = readUsableOllamaShowApiKey({
|
||||
env: process.env,
|
||||
allowAmbientEnvFallback: !isLocalOllamaBaseUrl(baseUrl),
|
||||
explicitApiKey: configuredApiKey,
|
||||
});
|
||||
// Managed secrets resolve asynchronously; retain their resolved fingerprint
|
||||
// per config so synchronous lookups cannot cross secret-provider ownership.
|
||||
const managedCredentialFingerprint =
|
||||
managedSecretScope && config
|
||||
? dynamicManagedCredentialFingerprints.get(config)?.get(managedSecretScope)
|
||||
: undefined;
|
||||
const credentialScope =
|
||||
apiKey ?? (secretRef ? `${secretRef.source}\0${secretRef.provider}\0${secretRef.id}` : "");
|
||||
const credentialFingerprint =
|
||||
managedCredentialFingerprint ?? createHash("sha256").update(credentialScope).digest("hex");
|
||||
return `${provider}\0${resolveOllamaApiBase(baseUrl)}\0${credentialFingerprint}`;
|
||||
}
|
||||
|
||||
function hasOllamaDiscoverySignal(providerConfig: ModelProviderConfig | undefined): boolean {
|
||||
return (
|
||||
Boolean(process.env.OLLAMA_API_KEY?.trim()) ||
|
||||
@@ -1173,20 +1132,9 @@ export default definePluginEntry({
|
||||
providerId: ctx.provider,
|
||||
});
|
||||
if (!hasOllamaDiscoverySignal(providerConfig)) {
|
||||
return;
|
||||
return undefined;
|
||||
}
|
||||
const baseUrl = readProviderBaseUrl(providerConfig);
|
||||
const managedSecretScope = buildDynamicManagedSecretScope(
|
||||
ctx.provider,
|
||||
baseUrl,
|
||||
providerConfig?.apiKey,
|
||||
);
|
||||
let dynamicCacheKey = buildDynamicCacheKey(
|
||||
ctx.provider,
|
||||
baseUrl,
|
||||
providerConfig?.apiKey,
|
||||
ctx.config,
|
||||
);
|
||||
let discoveryApiKey: string | undefined;
|
||||
if (providerConfig?.apiKey !== undefined && providerConfig.apiKey !== null) {
|
||||
const resolved = await resolveConfiguredSecretInputString({
|
||||
@@ -1197,11 +1145,7 @@ export default definePluginEntry({
|
||||
unresolvedReasonStyle: "detailed",
|
||||
});
|
||||
if (resolved.unresolvedRefReason) {
|
||||
dynamicModelCache.delete(dynamicCacheKey);
|
||||
if (managedSecretScope && ctx.config) {
|
||||
dynamicManagedCredentialFingerprints.get(ctx.config)?.delete(managedSecretScope);
|
||||
}
|
||||
return;
|
||||
return undefined;
|
||||
}
|
||||
const resolvedApiKey = readConfiguredOllamaApiKey(resolved.value);
|
||||
const configuredSecretRef = coerceSecretRef(providerConfig.apiKey);
|
||||
@@ -1211,38 +1155,11 @@ export default definePluginEntry({
|
||||
? readConcreteOllamaApiKey(process.env.OLLAMA_API_KEY)
|
||||
: readConcreteOllamaApiKey(resolvedApiKey);
|
||||
if (configuredSecretRef && !discoveryApiKey) {
|
||||
dynamicModelCache.delete(dynamicCacheKey);
|
||||
if (managedSecretScope && ctx.config) {
|
||||
dynamicManagedCredentialFingerprints.get(ctx.config)?.delete(managedSecretScope);
|
||||
}
|
||||
return;
|
||||
return undefined;
|
||||
}
|
||||
} else if (!isLocalOllamaBaseUrl(baseUrl)) {
|
||||
discoveryApiKey = readConcreteOllamaApiKey(process.env.OLLAMA_API_KEY);
|
||||
}
|
||||
if (managedSecretScope && ctx.config && discoveryApiKey) {
|
||||
let fingerprints = dynamicManagedCredentialFingerprints.get(ctx.config);
|
||||
if (!fingerprints) {
|
||||
fingerprints = new Map();
|
||||
dynamicManagedCredentialFingerprints.set(ctx.config, fingerprints);
|
||||
}
|
||||
const resolvedCredentialFingerprint = createHash("sha256")
|
||||
.update(discoveryApiKey)
|
||||
.digest("hex");
|
||||
if (
|
||||
fingerprints.has(managedSecretScope) &&
|
||||
fingerprints.get(managedSecretScope) !== resolvedCredentialFingerprint
|
||||
) {
|
||||
dynamicModelCache.delete(dynamicCacheKey);
|
||||
}
|
||||
fingerprints.set(managedSecretScope, resolvedCredentialFingerprint);
|
||||
dynamicCacheKey = buildDynamicCacheKey(
|
||||
ctx.provider,
|
||||
baseUrl,
|
||||
providerConfig?.apiKey,
|
||||
ctx.config,
|
||||
);
|
||||
}
|
||||
const provider = await buildLocalOllamaProvider(baseUrl, {
|
||||
quiet: true,
|
||||
...(discoveryApiKey ? { apiKey: discoveryApiKey } : {}),
|
||||
@@ -1257,42 +1174,21 @@ export default definePluginEntry({
|
||||
}),
|
||||
api: dynamicApi,
|
||||
};
|
||||
const dynamicModels = (dynamicProvider.models ?? []).map((model) =>
|
||||
toDynamicOllamaModel({
|
||||
const discoveredModel = dynamicProvider.models?.find((model) => model.id === ctx.modelId);
|
||||
if (discoveredModel) {
|
||||
return toDynamicOllamaModel({
|
||||
provider: ctx.provider,
|
||||
providerConfig: dynamicProvider,
|
||||
model,
|
||||
}),
|
||||
);
|
||||
if (!dynamicModels.some((model) => model.id === ctx.modelId)) {
|
||||
const requestedModel = await resolveRequestedDynamicOllamaModel({
|
||||
provider: ctx.provider,
|
||||
providerConfig: dynamicProvider,
|
||||
modelId: ctx.modelId,
|
||||
showApiKey: discoveryApiKey,
|
||||
capContextTokens: true,
|
||||
model: discoveredModel,
|
||||
});
|
||||
if (requestedModel) {
|
||||
dynamicModels.push(requestedModel);
|
||||
}
|
||||
}
|
||||
dynamicModelCache.set(dynamicCacheKey, dynamicModels);
|
||||
},
|
||||
resolveDynamicModel: (ctx) => {
|
||||
const providerConfig = resolveConfiguredOllamaProviderConfig({
|
||||
config: ctx.config,
|
||||
providerId: ctx.provider,
|
||||
return await resolveRequestedDynamicOllamaModel({
|
||||
provider: ctx.provider,
|
||||
providerConfig: dynamicProvider,
|
||||
modelId: ctx.modelId,
|
||||
showApiKey: discoveryApiKey,
|
||||
capContextTokens: true,
|
||||
});
|
||||
return dynamicModelCache
|
||||
.get(
|
||||
buildDynamicCacheKey(
|
||||
ctx.provider,
|
||||
readProviderBaseUrl(providerConfig),
|
||||
providerConfig?.apiKey,
|
||||
ctx.config,
|
||||
),
|
||||
)
|
||||
?.find((model) => model.id === ctx.modelId);
|
||||
},
|
||||
buildUnknownModelHint: () =>
|
||||
"Ollama requires authentication to be registered as a provider. " +
|
||||
|
||||
@@ -29,7 +29,7 @@ export type ProviderRuntimeHooks = {
|
||||
) => string | undefined;
|
||||
prepareProviderDynamicModel: (
|
||||
params: Parameters<typeof prepareProviderDynamicModel>[0],
|
||||
) => Promise<void>;
|
||||
) => ReturnType<typeof prepareProviderDynamicModel>;
|
||||
runProviderDynamicModel: (params: Parameters<typeof runProviderDynamicModel>[0]) => unknown;
|
||||
shouldPreferProviderRuntimeResolvedModel?: (
|
||||
params: Parameters<typeof shouldPreferProviderRuntimeResolvedModel>[0],
|
||||
|
||||
@@ -2,6 +2,7 @@ import type { OpenClawConfig } from "../../config/types.openclaw.js";
|
||||
import type { ModelRegistry as CoreModelRegistry } from "../../llm/model-registry.js";
|
||||
import type { Model } from "../../llm/types.js";
|
||||
import type { PluginMetadataSnapshotOwnerMaps } from "../../plugins/plugin-metadata-snapshot.types.js";
|
||||
import type { ProviderRuntimeModel } from "../../plugins/provider-runtime-model.types.js";
|
||||
import { ensureAuthProfileStore, resolveAuthProfileOrder } from "../auth-profiles.js";
|
||||
import type { AuthProfileCredential } from "../auth-profiles/types.js";
|
||||
import { resolveAgentHarnessPolicy } from "../harness/policy.js";
|
||||
@@ -252,21 +253,9 @@ export function resolveDynamicModelAuthProfile(params: {
|
||||
};
|
||||
}
|
||||
|
||||
function resolvePluginDynamicModelWithRegistry(params: {
|
||||
provider: string;
|
||||
modelId: string;
|
||||
modelRegistry: CoreModelRegistry;
|
||||
cfg?: OpenClawConfig;
|
||||
agentDir?: string;
|
||||
agentRuntimeId?: string;
|
||||
manifestAlias: ManifestModelCatalogProviderAliasMetadata;
|
||||
workspaceDir?: string;
|
||||
authProfileId?: string;
|
||||
authProfileMode?: AuthProfileCredential["type"] | "aws-sdk";
|
||||
preferredProfile?: string;
|
||||
runtimeHooks?: ProviderRuntimeHooks;
|
||||
getStaticCatalogModel?: () => StaticCatalogFallbackModel | undefined;
|
||||
}): Model | undefined {
|
||||
function resolvePluginDynamicModelWithRegistry(
|
||||
params: ResolveModelWithPreparedRegistryParams,
|
||||
): Model | undefined {
|
||||
const { provider, modelId, modelRegistry, cfg, agentDir, workspaceDir } = params;
|
||||
const runtimeHooks = params.runtimeHooks ?? DEFAULT_PROVIDER_RUNTIME_HOOKS;
|
||||
const providerConfig = resolveConfiguredProviderConfig(cfg, provider);
|
||||
@@ -294,22 +283,24 @@ function resolvePluginDynamicModelWithRegistry(params: {
|
||||
workspaceDir,
|
||||
runtimeHooks,
|
||||
});
|
||||
const pluginDynamicModel = runtimeHooks.runProviderDynamicModel({
|
||||
provider,
|
||||
config: cfg,
|
||||
workspaceDir,
|
||||
context: {
|
||||
config: cfg,
|
||||
agentDir,
|
||||
workspaceDir,
|
||||
...(agentRuntimeId ? { agentRuntimeId } : {}),
|
||||
const pluginDynamicModel =
|
||||
params.preparedDynamicModel ??
|
||||
(runtimeHooks.runProviderDynamicModel({
|
||||
provider,
|
||||
modelId,
|
||||
modelRegistry,
|
||||
providerConfig,
|
||||
...authProfile,
|
||||
},
|
||||
}) as Model | undefined;
|
||||
config: cfg,
|
||||
workspaceDir,
|
||||
context: {
|
||||
config: cfg,
|
||||
agentDir,
|
||||
workspaceDir,
|
||||
...(agentRuntimeId ? { agentRuntimeId } : {}),
|
||||
provider,
|
||||
modelId,
|
||||
modelRegistry,
|
||||
providerConfig,
|
||||
...authProfile,
|
||||
},
|
||||
}) as ProviderRuntimeModel | undefined);
|
||||
if (!pluginDynamicModel) {
|
||||
return undefined;
|
||||
}
|
||||
@@ -425,6 +416,7 @@ type ResolveModelWithRegistryParams = {
|
||||
|
||||
type ResolveModelWithPreparedRegistryParams = ResolveModelWithRegistryParams & {
|
||||
manifestAlias: ManifestModelCatalogProviderAliasMetadata;
|
||||
preparedDynamicModel?: ProviderRuntimeModel;
|
||||
getStaticCatalogModel?: () => StaticCatalogFallbackModel | undefined;
|
||||
};
|
||||
|
||||
|
||||
@@ -582,6 +582,127 @@ function makeVllmQwenConfig(
|
||||
}
|
||||
|
||||
describe("resolveModel", () => {
|
||||
it("consumes a directly prepared model through configured overrides and normalization", async () => {
|
||||
const preparedModel = {
|
||||
...makeModel("prepared-model"),
|
||||
provider: "acme",
|
||||
name: "Prepared Model",
|
||||
api: "openai-completions" as const,
|
||||
baseUrl: "https://discovered.example/v1",
|
||||
input: ["text" as const],
|
||||
contextWindow: 65_536,
|
||||
maxTokens: 8_192,
|
||||
};
|
||||
const prepareProviderDynamicModel = vi.fn(async () => preparedModel);
|
||||
const runProviderDynamicModel = vi.fn(() => undefined);
|
||||
const normalizeProviderResolvedModelWithPlugin = vi.fn(
|
||||
({ context }: { context: { model: Model } }) => ({
|
||||
...context.model,
|
||||
name: "Normalized Prepared Model",
|
||||
}),
|
||||
);
|
||||
const cfg = makeProviderConfig("acme", {
|
||||
api: "openai-responses",
|
||||
baseUrl: "https://configured.example/v1",
|
||||
headers: { "X-Tenant": "tenant-a" },
|
||||
});
|
||||
|
||||
const result = await resolveModelAsync("acme", "prepared-model", "/tmp/agent", cfg, {
|
||||
runtimeHooks: {
|
||||
...createRuntimeHooks(),
|
||||
prepareProviderDynamicModel,
|
||||
runProviderDynamicModel,
|
||||
normalizeProviderResolvedModelWithPlugin,
|
||||
},
|
||||
skipAgentDiscovery: true,
|
||||
});
|
||||
|
||||
expectRecordFields(expectResolvedModel(result), {
|
||||
provider: "acme",
|
||||
id: "prepared-model",
|
||||
name: "Normalized Prepared Model",
|
||||
api: "openai-responses",
|
||||
baseUrl: "https://configured.example/v1",
|
||||
contextWindow: 65_536,
|
||||
maxTokens: 8_192,
|
||||
});
|
||||
expect(expectResolvedModel(result).headers).toEqual(
|
||||
expect.objectContaining({ "X-Tenant": "tenant-a" }),
|
||||
);
|
||||
expect(prepareProviderDynamicModel).toHaveBeenCalledOnce();
|
||||
expect(normalizeProviderResolvedModelWithPlugin).toHaveBeenCalledOnce();
|
||||
expect(runProviderDynamicModel).not.toHaveBeenCalled();
|
||||
});
|
||||
|
||||
it.each([
|
||||
{
|
||||
description: "keeps explicit configured models ahead of prepared models",
|
||||
preferRuntime: false,
|
||||
expectedName: "Configured Model",
|
||||
expectedPreparationCount: 0,
|
||||
},
|
||||
{
|
||||
description: "preserves manually configured limits during runtime comparison",
|
||||
preferRuntime: true,
|
||||
expectedName: "Prepared Model",
|
||||
expectedPreparationCount: 1,
|
||||
},
|
||||
{
|
||||
description: "replaces models-add metadata with a preferred prepared model",
|
||||
preferRuntime: true,
|
||||
metadataSource: "models-add" as const,
|
||||
expectedName: "Prepared Model",
|
||||
expectedPreparationCount: 1,
|
||||
},
|
||||
])(
|
||||
"$description",
|
||||
async ({ preferRuntime, metadataSource, expectedName, expectedPreparationCount }) => {
|
||||
const prepareProviderDynamicModel = vi.fn(async () => ({
|
||||
...makeModel("prepared-model"),
|
||||
provider: "acme",
|
||||
name: "Prepared Model",
|
||||
api: "openai-completions" as const,
|
||||
baseUrl: "https://discovered.example/v1",
|
||||
input: ["text" as const],
|
||||
contextWindow: 65_536,
|
||||
maxTokens: 8_192,
|
||||
}));
|
||||
const runProviderDynamicModel = vi.fn(() => undefined);
|
||||
const cfg = makeProviderConfig("acme", {
|
||||
api: "openai-responses",
|
||||
baseUrl: "https://configured.example/v1",
|
||||
models: [
|
||||
{
|
||||
...makeModel("prepared-model"),
|
||||
name: "Configured Model",
|
||||
contextWindow: 32_768,
|
||||
maxTokens: 4_096,
|
||||
...(metadataSource ? { metadataSource } : {}),
|
||||
},
|
||||
],
|
||||
});
|
||||
|
||||
const result = await resolveModelAsync("acme", "prepared-model", "/tmp/agent", cfg, {
|
||||
runtimeHooks: {
|
||||
...createRuntimeHooks(),
|
||||
prepareProviderDynamicModel,
|
||||
runProviderDynamicModel,
|
||||
shouldPreferProviderRuntimeResolvedModel: () => preferRuntime,
|
||||
},
|
||||
skipAgentDiscovery: true,
|
||||
});
|
||||
|
||||
expectRecordFields(expectResolvedModel(result), {
|
||||
name: expectedName,
|
||||
api: "openai-responses",
|
||||
baseUrl: "https://configured.example/v1",
|
||||
contextWindow: metadataSource ? 65_536 : 32_768,
|
||||
});
|
||||
expect(prepareProviderDynamicModel).toHaveBeenCalledTimes(expectedPreparationCount);
|
||||
expect(runProviderDynamicModel).not.toHaveBeenCalled();
|
||||
},
|
||||
);
|
||||
|
||||
it("reuses agent discovery stores while the agent model files are unchanged", async () => {
|
||||
mockModelDiscovery();
|
||||
|
||||
|
||||
@@ -419,7 +419,7 @@ export async function resolveModelAsync(
|
||||
});
|
||||
};
|
||||
const resolveDynamicAttempt = async () => {
|
||||
await runtimeHooks.prepareProviderDynamicModel({
|
||||
const preparedDynamicModel = await runtimeHooks.prepareProviderDynamicModel({
|
||||
provider: normalizedRef.provider,
|
||||
config: cfg,
|
||||
workspaceDir,
|
||||
@@ -448,6 +448,7 @@ export async function resolveModelAsync(
|
||||
authProfileMode: options?.authProfileMode,
|
||||
preferredProfile: options?.preferredProfile,
|
||||
runtimeHooks,
|
||||
...(preparedDynamicModel ? { preparedDynamicModel } : {}),
|
||||
getStaticCatalogModel: getManifestStaticCatalogModel,
|
||||
...(options?.allowBundledStaticCatalogFallback ? { skipConfiguredFallback: true } : {}),
|
||||
});
|
||||
|
||||
@@ -155,6 +155,27 @@ describe("appendPrioritizedDynamicLiveModels", () => {
|
||||
expect(resolveDynamicModel).not.toHaveBeenCalled();
|
||||
});
|
||||
|
||||
it("materializes a directly prepared model without retrying synchronous resolution", async () => {
|
||||
const preparedModel = model(DYNAMIC_PROVIDER, "glm-5");
|
||||
providerRuntimeMocks.prepareProviderDynamicModel.mockResolvedValue(preparedModel);
|
||||
|
||||
const result = await appendPrioritizedDynamicLiveModels({
|
||||
models: [],
|
||||
agentDir: "/tmp/openclaw-agent",
|
||||
modelRegistry: REGISTRY,
|
||||
refs: [{ provider: DYNAMIC_PROVIDER, id: "glm-5" }],
|
||||
});
|
||||
|
||||
expect(result.added).toEqual([preparedModel]);
|
||||
expect(providerRuntimeMocks.prepareProviderDynamicModel).toHaveBeenCalledOnce();
|
||||
expect(providerRuntimeMocks.runProviderDynamicModel).not.toHaveBeenCalled();
|
||||
expect(normalizeDiscoveredAgentModelMock).toHaveBeenCalledWith(
|
||||
preparedModel,
|
||||
"/tmp/openclaw-agent",
|
||||
{ config: undefined, workspaceDir: undefined },
|
||||
);
|
||||
});
|
||||
|
||||
it("uses default provider runtime hooks when resolvers are not injected", async () => {
|
||||
providerRuntimeMocks.runProviderDynamicModel.mockImplementation((params) =>
|
||||
params.context.provider === DYNAMIC_PROVIDER && params.context.modelId === "glm-5"
|
||||
|
||||
@@ -29,9 +29,9 @@ const providerRuntimeLoader = createLazyImportLoader<ProviderRuntimeModule>(
|
||||
|
||||
async function prepareProviderDynamicModelDefault(
|
||||
params: Parameters<DynamicModelPreparer>[0],
|
||||
): Promise<void> {
|
||||
): ReturnType<DynamicModelPreparer> {
|
||||
const { prepareProviderDynamicModel } = await providerRuntimeLoader.load();
|
||||
await prepareProviderDynamicModel(params);
|
||||
return await prepareProviderDynamicModel(params);
|
||||
}
|
||||
|
||||
async function runProviderDynamicModelDefault(
|
||||
@@ -108,20 +108,22 @@ export async function appendPrioritizedDynamicLiveModels(params: {
|
||||
modelRegistry: params.modelRegistry,
|
||||
providerConfig,
|
||||
};
|
||||
await prepareDynamicModel({
|
||||
provider: ref.provider,
|
||||
config: params.config,
|
||||
workspaceDir: params.workspaceDir,
|
||||
env: params.env,
|
||||
context,
|
||||
});
|
||||
const resolved = await resolveDynamicModel({
|
||||
const prepared = await prepareDynamicModel({
|
||||
provider: ref.provider,
|
||||
config: params.config,
|
||||
workspaceDir: params.workspaceDir,
|
||||
env: params.env,
|
||||
context,
|
||||
});
|
||||
const resolved =
|
||||
prepared ??
|
||||
(await resolveDynamicModel({
|
||||
provider: ref.provider,
|
||||
config: params.config,
|
||||
workspaceDir: params.workspaceDir,
|
||||
env: params.env,
|
||||
context,
|
||||
}));
|
||||
if (!resolved) {
|
||||
continue;
|
||||
}
|
||||
|
||||
@@ -154,19 +154,21 @@ export type ProviderPlugin = {
|
||||
* 3. core fallback heuristics
|
||||
* 4. generic provider-config fallback
|
||||
*
|
||||
* Keep this hook cheap and deterministic. If you need network I/O first, use
|
||||
* `prepareDynamicModel` to prime state for the async retry path.
|
||||
* Keep this hook cheap and deterministic. Async model discovery belongs in
|
||||
* `prepareDynamicModel`, which can return the prepared model directly.
|
||||
*/
|
||||
resolveDynamicModel?: (
|
||||
ctx: ProviderResolveDynamicModelContext,
|
||||
) => ProviderRuntimeModel | null | undefined;
|
||||
/**
|
||||
* Optional async prefetch for dynamic model resolution.
|
||||
* Optional async preparation for dynamic model resolution.
|
||||
*
|
||||
* OpenClaw calls this only from async model resolution paths. After it
|
||||
* completes, `resolveDynamicModel` is called again.
|
||||
* OpenClaw calls this only from async model resolution paths. Return the
|
||||
* requested model directly, or return nothing to retry `resolveDynamicModel`.
|
||||
*/
|
||||
prepareDynamicModel?: (ctx: ProviderPrepareDynamicModelContext) => Promise<void>;
|
||||
prepareDynamicModel?: (
|
||||
ctx: ProviderPrepareDynamicModelContext,
|
||||
) => Promise<ProviderRuntimeModel | void>;
|
||||
/**
|
||||
* Lets a provider plugin opt exact configured models into a runtime
|
||||
* metadata comparison pass before the embedded runner returns the explicit
|
||||
|
||||
@@ -2041,7 +2041,7 @@ describe("provider-runtime", () => {
|
||||
{ pluginId: "demo", providerIds: ["demo"] },
|
||||
]);
|
||||
resolveExternalAuthProfileProviderPluginIdsMock.mockReturnValue(["demo"]);
|
||||
const prepareDynamicModel = vi.fn(async () => undefined);
|
||||
const prepareDynamicModel = vi.fn(async () => MODEL);
|
||||
const createStreamFn = vi.fn(() => vi.fn());
|
||||
const sanitizeReplayHistory = vi.fn(
|
||||
async ({
|
||||
@@ -2245,12 +2245,14 @@ describe("provider-runtime", () => {
|
||||
}),
|
||||
).toBe("DEMO_PROFILE");
|
||||
|
||||
await prepareProviderDynamicModel({
|
||||
provider: DEMO_PROVIDER_ID,
|
||||
context: createDemoRuntimeContext({
|
||||
modelRegistry: EMPTY_MODEL_REGISTRY,
|
||||
expect(
|
||||
await prepareProviderDynamicModel({
|
||||
provider: DEMO_PROVIDER_ID,
|
||||
context: createDemoRuntimeContext({
|
||||
modelRegistry: EMPTY_MODEL_REGISTRY,
|
||||
}),
|
||||
}),
|
||||
});
|
||||
).toEqual(MODEL);
|
||||
|
||||
expect(
|
||||
resolveProviderReasoningOutputModeWithPlugin({
|
||||
|
||||
@@ -282,8 +282,8 @@ export async function prepareProviderDynamicModel(params: {
|
||||
workspaceDir?: string;
|
||||
env?: NodeJS.ProcessEnv;
|
||||
context: ProviderPrepareDynamicModelContext;
|
||||
}): Promise<void> {
|
||||
await resolveProviderRuntimePlugin(params)?.prepareDynamicModel?.(params.context);
|
||||
}): Promise<ProviderRuntimeModel | void> {
|
||||
return resolveProviderRuntimePlugin(params)?.prepareDynamicModel?.(params.context);
|
||||
}
|
||||
|
||||
export function shouldPreferProviderRuntimeResolvedModel(params: {
|
||||
|
||||
@@ -41,11 +41,10 @@ export type ProviderResolveDynamicModelContext = {
|
||||
};
|
||||
|
||||
/**
|
||||
* Optional async warm-up for dynamic model resolution.
|
||||
* Optional async preparation for dynamic model resolution.
|
||||
*
|
||||
* Called only from async model resolution paths, before retrying
|
||||
* `resolveDynamicModel`. This is the place to refresh caches or fetch provider
|
||||
* metadata over the network.
|
||||
* Called only from async model resolution paths. Providers can return the
|
||||
* requested model directly or refresh reusable metadata before the sync retry.
|
||||
*/
|
||||
export type ProviderPrepareDynamicModelContext = ProviderResolveDynamicModelContext;
|
||||
|
||||
|
||||
Reference in New Issue
Block a user