mirror of
https://github.com/openclaw/openclaw.git
synced 2026-05-02 07:50:22 +00:00
refactor: move provider runtime into extensions
This commit is contained in:
Peter Steinberger
@@ -1,5 +1,4 @@
|
||||
import { normalizeProviderId } from "../agents/provider-id.js";
|
||||
import { findCatalogTemplate } from "./provider-catalog.js";
|
||||
import type {
|
||||
ProviderAugmentModelCatalogContext,
|
||||
ProviderBuiltInModelSuppressionContext,
|
||||
@@ -10,6 +9,22 @@ const OPENAI_CODEX_PROVIDER_ID = "openai-codex";
|
||||
const OPENAI_DIRECT_SPARK_MODEL_ID = "gpt-5.3-codex-spark";
|
||||
const SUPPRESSED_SPARK_PROVIDERS = new Set(["openai", "azure-openai-responses"]);
|
||||
|
||||
function findCatalogTemplate(params: {
|
||||
entries: ReadonlyArray<{ provider: string; id: string }>;
|
||||
providerId: string;
|
||||
templateIds: readonly string[];
|
||||
}) {
|
||||
return params.templateIds
|
||||
.map((templateId) =>
|
||||
params.entries.find(
|
||||
(entry) =>
|
||||
entry.provider.toLowerCase() === params.providerId.toLowerCase() &&
|
||||
entry.id.toLowerCase() === templateId.toLowerCase(),
|
||||
),
|
||||
)
|
||||
.find((entry) => entry !== undefined);
|
||||
}
|
||||
|
||||
export function resolveBundledProviderBuiltInModelSuppression(
|
||||
context: ProviderBuiltInModelSuppressionContext,
|
||||
) {
|
||||
|
||||
@@ -1,533 +1,8 @@
|
||||
import { upsertAuthProfileWithLock } from "../agents/auth-profiles/upsert-with-lock.js";
|
||||
import { OLLAMA_DEFAULT_BASE_URL } from "../agents/ollama-defaults.js";
|
||||
import {
|
||||
buildOllamaModelDefinition,
|
||||
enrichOllamaModelsWithContext,
|
||||
fetchOllamaModels,
|
||||
resolveOllamaApiBase,
|
||||
type OllamaModelWithContext,
|
||||
} from "../agents/ollama-models.js";
|
||||
import type { OpenClawConfig } from "../config/config.js";
|
||||
import type { RuntimeEnv } from "../runtime.js";
|
||||
import { WizardCancelledError, type WizardPrompter } from "../wizard/prompts.js";
|
||||
import { applyAgentDefaultModelPrimary } from "./provider-onboarding-config.js";
|
||||
import { isRemoteEnvironment, openUrl } from "./setup-browser.js";
|
||||
import type { ProviderAuthOptionBag } from "./types.js";
|
||||
|
||||
export { OLLAMA_DEFAULT_BASE_URL } from "../agents/ollama-defaults.js";
|
||||
export const OLLAMA_DEFAULT_MODEL = "glm-4.7-flash";
|
||||
|
||||
const OLLAMA_SUGGESTED_MODELS_LOCAL = ["glm-4.7-flash"];
|
||||
const OLLAMA_SUGGESTED_MODELS_CLOUD = ["kimi-k2.5:cloud", "minimax-m2.5:cloud", "glm-5:cloud"];
|
||||
type OllamaMode = "remote" | "local";
|
||||
type OllamaSetupOptions = ProviderAuthOptionBag & {
|
||||
customBaseUrl?: string;
|
||||
customModelId?: string;
|
||||
};
|
||||
|
||||
function normalizeOllamaModelName(value: string | undefined): string | undefined {
|
||||
const trimmed = value?.trim();
|
||||
if (!trimmed) {
|
||||
return undefined;
|
||||
}
|
||||
if (trimmed.toLowerCase().startsWith("ollama/")) {
|
||||
const withoutPrefix = trimmed.slice("ollama/".length).trim();
|
||||
return withoutPrefix || undefined;
|
||||
}
|
||||
return trimmed;
|
||||
}
|
||||
|
||||
function isOllamaCloudModel(modelName: string | undefined): boolean {
|
||||
return Boolean(modelName?.trim().toLowerCase().endsWith(":cloud"));
|
||||
}
|
||||
|
||||
function formatOllamaPullStatus(status: string): { text: string; hidePercent: boolean } {
|
||||
const trimmed = status.trim();
|
||||
const partStatusMatch = trimmed.match(/^([a-z-]+)\s+(?:sha256:)?[a-f0-9]{8,}$/i);
|
||||
if (partStatusMatch) {
|
||||
return { text: `${partStatusMatch[1]} part`, hidePercent: false };
|
||||
}
|
||||
if (/^verifying\b.*\bdigest\b/i.test(trimmed)) {
|
||||
return { text: "verifying digest", hidePercent: true };
|
||||
}
|
||||
return { text: trimmed, hidePercent: false };
|
||||
}
|
||||
|
||||
type OllamaCloudAuthResult = {
|
||||
signedIn: boolean;
|
||||
signinUrl?: string;
|
||||
};
|
||||
|
||||
/** Check if the user is signed in to Ollama cloud via /api/me. */
|
||||
async function checkOllamaCloudAuth(baseUrl: string): Promise<OllamaCloudAuthResult> {
|
||||
try {
|
||||
const apiBase = resolveOllamaApiBase(baseUrl);
|
||||
const response = await fetch(`${apiBase}/api/me`, {
|
||||
method: "POST",
|
||||
signal: AbortSignal.timeout(5000),
|
||||
});
|
||||
if (response.status === 401) {
|
||||
// 401 body contains { error, signin_url }
|
||||
const data = (await response.json()) as { signin_url?: string };
|
||||
return { signedIn: false, signinUrl: data.signin_url };
|
||||
}
|
||||
if (!response.ok) {
|
||||
return { signedIn: false };
|
||||
}
|
||||
return { signedIn: true };
|
||||
} catch {
|
||||
// /api/me not supported or unreachable — fail closed so cloud mode
|
||||
// doesn't silently skip auth; the caller handles the fallback.
|
||||
return { signedIn: false };
|
||||
}
|
||||
}
|
||||
|
||||
type OllamaPullChunk = {
|
||||
status?: string;
|
||||
total?: number;
|
||||
completed?: number;
|
||||
error?: string;
|
||||
};
|
||||
|
||||
type OllamaPullFailureKind = "http" | "no-body" | "chunk-error" | "network";
|
||||
type OllamaPullResult =
|
||||
| { ok: true }
|
||||
| {
|
||||
ok: false;
|
||||
kind: OllamaPullFailureKind;
|
||||
message: string;
|
||||
};
|
||||
|
||||
async function pullOllamaModelCore(params: {
|
||||
baseUrl: string;
|
||||
modelName: string;
|
||||
onStatus?: (status: string, percent: number | null) => void;
|
||||
}): Promise<OllamaPullResult> {
|
||||
const { onStatus } = params;
|
||||
const baseUrl = resolveOllamaApiBase(params.baseUrl);
|
||||
const modelName = normalizeOllamaModelName(params.modelName) ?? params.modelName.trim();
|
||||
try {
|
||||
const response = await fetch(`${baseUrl}/api/pull`, {
|
||||
method: "POST",
|
||||
headers: { "Content-Type": "application/json" },
|
||||
body: JSON.stringify({ name: modelName }),
|
||||
});
|
||||
if (!response.ok) {
|
||||
return {
|
||||
ok: false,
|
||||
kind: "http",
|
||||
message: `Failed to download ${modelName} (HTTP ${response.status})`,
|
||||
};
|
||||
}
|
||||
if (!response.body) {
|
||||
return {
|
||||
ok: false,
|
||||
kind: "no-body",
|
||||
message: `Failed to download ${modelName} (no response body)`,
|
||||
};
|
||||
}
|
||||
|
||||
const reader = response.body.getReader();
|
||||
const decoder = new TextDecoder();
|
||||
let buffer = "";
|
||||
const layers = new Map<string, { total: number; completed: number }>();
|
||||
|
||||
const parseLine = (line: string): OllamaPullResult => {
|
||||
const trimmed = line.trim();
|
||||
if (!trimmed) {
|
||||
return { ok: true };
|
||||
}
|
||||
try {
|
||||
const chunk = JSON.parse(trimmed) as OllamaPullChunk;
|
||||
if (chunk.error) {
|
||||
return {
|
||||
ok: false,
|
||||
kind: "chunk-error",
|
||||
message: `Download failed: ${chunk.error}`,
|
||||
};
|
||||
}
|
||||
if (!chunk.status) {
|
||||
return { ok: true };
|
||||
}
|
||||
if (chunk.total && chunk.completed !== undefined) {
|
||||
layers.set(chunk.status, { total: chunk.total, completed: chunk.completed });
|
||||
let totalSum = 0;
|
||||
let completedSum = 0;
|
||||
for (const layer of layers.values()) {
|
||||
totalSum += layer.total;
|
||||
completedSum += layer.completed;
|
||||
}
|
||||
const percent = totalSum > 0 ? Math.round((completedSum / totalSum) * 100) : null;
|
||||
onStatus?.(chunk.status, percent);
|
||||
} else {
|
||||
onStatus?.(chunk.status, null);
|
||||
}
|
||||
} catch {
|
||||
// Ignore malformed lines from streaming output.
|
||||
}
|
||||
return { ok: true };
|
||||
};
|
||||
|
||||
for (;;) {
|
||||
const { done, value } = await reader.read();
|
||||
if (done) {
|
||||
break;
|
||||
}
|
||||
buffer += decoder.decode(value, { stream: true });
|
||||
const lines = buffer.split("\n");
|
||||
buffer = lines.pop() ?? "";
|
||||
for (const line of lines) {
|
||||
const parsed = parseLine(line);
|
||||
if (!parsed.ok) {
|
||||
return parsed;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
const trailing = buffer.trim();
|
||||
if (trailing) {
|
||||
const parsed = parseLine(trailing);
|
||||
if (!parsed.ok) {
|
||||
return parsed;
|
||||
}
|
||||
}
|
||||
|
||||
return { ok: true };
|
||||
} catch (err) {
|
||||
const reason = err instanceof Error ? err.message : String(err);
|
||||
return {
|
||||
ok: false,
|
||||
kind: "network",
|
||||
message: `Failed to download ${modelName}: ${reason}`,
|
||||
};
|
||||
}
|
||||
}
|
||||
|
||||
/** Pull a model from Ollama, streaming progress updates. */
|
||||
async function pullOllamaModel(
|
||||
baseUrl: string,
|
||||
modelName: string,
|
||||
prompter: WizardPrompter,
|
||||
): Promise<boolean> {
|
||||
const spinner = prompter.progress(`Downloading ${modelName}...`);
|
||||
const result = await pullOllamaModelCore({
|
||||
baseUrl,
|
||||
modelName,
|
||||
onStatus: (status, percent) => {
|
||||
const displayStatus = formatOllamaPullStatus(status);
|
||||
if (displayStatus.hidePercent) {
|
||||
spinner.update(`Downloading ${modelName} - ${displayStatus.text}`);
|
||||
} else {
|
||||
spinner.update(`Downloading ${modelName} - ${displayStatus.text} - ${percent ?? 0}%`);
|
||||
}
|
||||
},
|
||||
});
|
||||
if (!result.ok) {
|
||||
spinner.stop(result.message);
|
||||
return false;
|
||||
}
|
||||
spinner.stop(`Downloaded ${modelName}`);
|
||||
return true;
|
||||
}
|
||||
|
||||
async function pullOllamaModelNonInteractive(
|
||||
baseUrl: string,
|
||||
modelName: string,
|
||||
runtime: RuntimeEnv,
|
||||
): Promise<boolean> {
|
||||
runtime.log(`Downloading ${modelName}...`);
|
||||
const result = await pullOllamaModelCore({ baseUrl, modelName });
|
||||
if (!result.ok) {
|
||||
runtime.error(result.message);
|
||||
return false;
|
||||
}
|
||||
runtime.log(`Downloaded ${modelName}`);
|
||||
return true;
|
||||
}
|
||||
|
||||
function buildOllamaModelsConfig(
|
||||
modelNames: string[],
|
||||
discoveredModelsByName?: Map<string, OllamaModelWithContext>,
|
||||
) {
|
||||
return modelNames.map((name) =>
|
||||
buildOllamaModelDefinition(name, discoveredModelsByName?.get(name)?.contextWindow),
|
||||
);
|
||||
}
|
||||
|
||||
function applyOllamaProviderConfig(
|
||||
cfg: OpenClawConfig,
|
||||
baseUrl: string,
|
||||
modelNames: string[],
|
||||
discoveredModelsByName?: Map<string, OllamaModelWithContext>,
|
||||
): OpenClawConfig {
|
||||
return {
|
||||
...cfg,
|
||||
models: {
|
||||
...cfg.models,
|
||||
mode: cfg.models?.mode ?? "merge",
|
||||
providers: {
|
||||
...cfg.models?.providers,
|
||||
ollama: {
|
||||
baseUrl,
|
||||
api: "ollama",
|
||||
apiKey: "OLLAMA_API_KEY", // pragma: allowlist secret
|
||||
models: buildOllamaModelsConfig(modelNames, discoveredModelsByName),
|
||||
},
|
||||
},
|
||||
},
|
||||
};
|
||||
}
|
||||
|
||||
async function storeOllamaCredential(agentDir?: string): Promise<void> {
|
||||
await upsertAuthProfileWithLock({
|
||||
profileId: "ollama:default",
|
||||
credential: { type: "api_key", provider: "ollama", key: "ollama-local" },
|
||||
agentDir,
|
||||
});
|
||||
}
|
||||
|
||||
/**
|
||||
* Interactive: prompt for base URL, discover models, configure provider.
|
||||
* Model selection is handled by the standard model picker downstream.
|
||||
*/
|
||||
export async function promptAndConfigureOllama(params: {
|
||||
cfg: OpenClawConfig;
|
||||
prompter: WizardPrompter;
|
||||
}): Promise<{ config: OpenClawConfig }> {
|
||||
const { prompter } = params;
|
||||
|
||||
// 1. Prompt base URL
|
||||
const baseUrlRaw = await prompter.text({
|
||||
message: "Ollama base URL",
|
||||
initialValue: OLLAMA_DEFAULT_BASE_URL,
|
||||
placeholder: OLLAMA_DEFAULT_BASE_URL,
|
||||
validate: (value) => (value?.trim() ? undefined : "Required"),
|
||||
});
|
||||
const configuredBaseUrl = String(baseUrlRaw ?? "")
|
||||
.trim()
|
||||
.replace(/\/+$/, "");
|
||||
const baseUrl = resolveOllamaApiBase(configuredBaseUrl);
|
||||
|
||||
// 2. Check reachability
|
||||
const { reachable, models } = await fetchOllamaModels(baseUrl);
|
||||
|
||||
if (!reachable) {
|
||||
await prompter.note(
|
||||
[
|
||||
`Ollama could not be reached at ${baseUrl}.`,
|
||||
"Download it at https://ollama.com/download",
|
||||
"",
|
||||
"Start Ollama and re-run setup.",
|
||||
].join("\n"),
|
||||
"Ollama",
|
||||
);
|
||||
throw new WizardCancelledError("Ollama not reachable");
|
||||
}
|
||||
|
||||
const enrichedModels = await enrichOllamaModelsWithContext(baseUrl, models.slice(0, 50));
|
||||
const discoveredModelsByName = new Map(enrichedModels.map((model) => [model.name, model]));
|
||||
const modelNames = models.map((m) => m.name);
|
||||
|
||||
// 3. Mode selection
|
||||
const mode = (await prompter.select({
|
||||
message: "Ollama mode",
|
||||
options: [
|
||||
{ value: "remote", label: "Cloud + Local", hint: "Ollama cloud models + local models" },
|
||||
{ value: "local", label: "Local", hint: "Local models only" },
|
||||
],
|
||||
})) as OllamaMode;
|
||||
|
||||
// 4. Cloud auth — check /api/me upfront for remote (cloud+local) mode
|
||||
let cloudAuthVerified = false;
|
||||
if (mode === "remote") {
|
||||
const authResult = await checkOllamaCloudAuth(baseUrl);
|
||||
if (!authResult.signedIn) {
|
||||
if (authResult.signinUrl) {
|
||||
if (!isRemoteEnvironment()) {
|
||||
await openUrl(authResult.signinUrl);
|
||||
}
|
||||
await prompter.note(
|
||||
["Sign in to Ollama Cloud:", authResult.signinUrl].join("\n"),
|
||||
"Ollama Cloud",
|
||||
);
|
||||
const confirmed = await prompter.confirm({
|
||||
message: "Have you signed in?",
|
||||
});
|
||||
if (!confirmed) {
|
||||
throw new WizardCancelledError("Ollama cloud sign-in cancelled");
|
||||
}
|
||||
// Re-check after user claims sign-in
|
||||
const recheck = await checkOllamaCloudAuth(baseUrl);
|
||||
if (!recheck.signedIn) {
|
||||
throw new WizardCancelledError("Ollama cloud sign-in required");
|
||||
}
|
||||
cloudAuthVerified = true;
|
||||
} else {
|
||||
// No signin URL available (older server, unreachable /api/me, or custom gateway).
|
||||
await prompter.note(
|
||||
[
|
||||
"Could not verify Ollama Cloud authentication.",
|
||||
"Cloud models may not work until you sign in at https://ollama.com.",
|
||||
].join("\n"),
|
||||
"Ollama Cloud",
|
||||
);
|
||||
const continueAnyway = await prompter.confirm({
|
||||
message: "Continue without cloud auth?",
|
||||
});
|
||||
if (!continueAnyway) {
|
||||
throw new WizardCancelledError("Ollama cloud auth could not be verified");
|
||||
}
|
||||
// Cloud auth unverified — fall back to local defaults so the model
|
||||
// picker doesn't steer toward cloud models that may fail.
|
||||
}
|
||||
} else {
|
||||
cloudAuthVerified = true;
|
||||
}
|
||||
}
|
||||
|
||||
// 5. Model ordering — suggested models first.
|
||||
// Use cloud defaults only when auth was actually verified; otherwise fall
|
||||
// back to local defaults so the user isn't steered toward cloud models
|
||||
// that may fail at runtime.
|
||||
const suggestedModels =
|
||||
mode === "local" || !cloudAuthVerified
|
||||
? OLLAMA_SUGGESTED_MODELS_LOCAL
|
||||
: OLLAMA_SUGGESTED_MODELS_CLOUD;
|
||||
const orderedModelNames = [
|
||||
...suggestedModels,
|
||||
...modelNames.filter((name) => !suggestedModels.includes(name)),
|
||||
];
|
||||
|
||||
const config = applyOllamaProviderConfig(
|
||||
params.cfg,
|
||||
baseUrl,
|
||||
orderedModelNames,
|
||||
discoveredModelsByName,
|
||||
);
|
||||
return { config };
|
||||
}
|
||||
|
||||
/** Non-interactive: auto-discover models and configure provider. */
|
||||
export async function configureOllamaNonInteractive(params: {
|
||||
nextConfig: OpenClawConfig;
|
||||
opts: OllamaSetupOptions;
|
||||
runtime: RuntimeEnv;
|
||||
}): Promise<OpenClawConfig> {
|
||||
const { opts, runtime } = params;
|
||||
const configuredBaseUrl = (opts.customBaseUrl?.trim() || OLLAMA_DEFAULT_BASE_URL).replace(
|
||||
/\/+$/,
|
||||
"",
|
||||
);
|
||||
const baseUrl = resolveOllamaApiBase(configuredBaseUrl);
|
||||
|
||||
const { reachable, models } = await fetchOllamaModels(baseUrl);
|
||||
const explicitModel = normalizeOllamaModelName(opts.customModelId);
|
||||
|
||||
if (!reachable) {
|
||||
runtime.error(
|
||||
[
|
||||
`Ollama could not be reached at ${baseUrl}.`,
|
||||
"Download it at https://ollama.com/download",
|
||||
].join("\n"),
|
||||
);
|
||||
runtime.exit(1);
|
||||
return params.nextConfig;
|
||||
}
|
||||
|
||||
await storeOllamaCredential();
|
||||
|
||||
const enrichedModels = await enrichOllamaModelsWithContext(baseUrl, models.slice(0, 50));
|
||||
const discoveredModelsByName = new Map(enrichedModels.map((model) => [model.name, model]));
|
||||
const modelNames = models.map((m) => m.name);
|
||||
|
||||
// Apply local suggested model ordering.
|
||||
const suggestedModels = OLLAMA_SUGGESTED_MODELS_LOCAL;
|
||||
const orderedModelNames = [
|
||||
...suggestedModels,
|
||||
...modelNames.filter((name) => !suggestedModels.includes(name)),
|
||||
];
|
||||
|
||||
const requestedDefaultModelId = explicitModel ?? suggestedModels[0];
|
||||
let pulledRequestedModel = false;
|
||||
const availableModelNames = new Set(modelNames);
|
||||
const requestedCloudModel = isOllamaCloudModel(requestedDefaultModelId);
|
||||
|
||||
if (requestedCloudModel) {
|
||||
availableModelNames.add(requestedDefaultModelId);
|
||||
}
|
||||
|
||||
// Pull if model not in discovered list and Ollama is reachable
|
||||
if (!requestedCloudModel && !modelNames.includes(requestedDefaultModelId)) {
|
||||
pulledRequestedModel = await pullOllamaModelNonInteractive(
|
||||
baseUrl,
|
||||
requestedDefaultModelId,
|
||||
runtime,
|
||||
);
|
||||
if (pulledRequestedModel) {
|
||||
availableModelNames.add(requestedDefaultModelId);
|
||||
}
|
||||
}
|
||||
|
||||
let allModelNames = orderedModelNames;
|
||||
let defaultModelId = requestedDefaultModelId;
|
||||
if (
|
||||
(pulledRequestedModel || requestedCloudModel) &&
|
||||
!allModelNames.includes(requestedDefaultModelId)
|
||||
) {
|
||||
allModelNames = [...allModelNames, requestedDefaultModelId];
|
||||
}
|
||||
if (!availableModelNames.has(requestedDefaultModelId)) {
|
||||
if (availableModelNames.size > 0) {
|
||||
const firstAvailableModel =
|
||||
allModelNames.find((name) => availableModelNames.has(name)) ??
|
||||
Array.from(availableModelNames)[0];
|
||||
defaultModelId = firstAvailableModel;
|
||||
runtime.log(
|
||||
`Ollama model ${requestedDefaultModelId} was not available; using ${defaultModelId} instead.`,
|
||||
);
|
||||
} else {
|
||||
runtime.error(
|
||||
[
|
||||
`No Ollama models are available at ${baseUrl}.`,
|
||||
"Pull a model first, then re-run setup.",
|
||||
].join("\n"),
|
||||
);
|
||||
runtime.exit(1);
|
||||
return params.nextConfig;
|
||||
}
|
||||
}
|
||||
|
||||
const config = applyOllamaProviderConfig(
|
||||
params.nextConfig,
|
||||
baseUrl,
|
||||
allModelNames,
|
||||
discoveredModelsByName,
|
||||
);
|
||||
const modelRef = `ollama/${defaultModelId}`;
|
||||
runtime.log(`Default Ollama model: ${defaultModelId}`);
|
||||
return applyAgentDefaultModelPrimary(config, modelRef);
|
||||
}
|
||||
|
||||
/** Pull the configured default Ollama model if it isn't already available locally. */
|
||||
export async function ensureOllamaModelPulled(params: {
|
||||
config: OpenClawConfig;
|
||||
model: string;
|
||||
prompter: WizardPrompter;
|
||||
}): Promise<void> {
|
||||
if (!params.model.startsWith("ollama/")) {
|
||||
return;
|
||||
}
|
||||
const baseUrl = params.config.models?.providers?.ollama?.baseUrl ?? OLLAMA_DEFAULT_BASE_URL;
|
||||
const modelName = params.model.slice("ollama/".length);
|
||||
if (isOllamaCloudModel(modelName)) {
|
||||
return;
|
||||
}
|
||||
const { models } = await fetchOllamaModels(baseUrl);
|
||||
if (models.some((m) => m.name === modelName)) {
|
||||
return;
|
||||
}
|
||||
const pulled = await pullOllamaModel(baseUrl, modelName, params.prompter);
|
||||
if (!pulled) {
|
||||
throw new WizardCancelledError("Failed to download selected Ollama model");
|
||||
}
|
||||
}
|
||||
export {
|
||||
OLLAMA_DEFAULT_BASE_URL,
|
||||
OLLAMA_DEFAULT_MODEL,
|
||||
buildOllamaProvider,
|
||||
configureOllamaNonInteractive,
|
||||
ensureOllamaModelPulled,
|
||||
promptAndConfigureOllama,
|
||||
} from "../plugin-sdk/provider-setup.js";
|
||||
|
||||
@@ -23,13 +23,17 @@ const resolveOwningPluginIdsForProviderMock = vi.fn<ResolveOwningPluginIdsForPro
|
||||
let augmentModelCatalogWithProviderPlugins: typeof import("./provider-runtime.js").augmentModelCatalogWithProviderPlugins;
|
||||
let buildProviderAuthDoctorHintWithPlugin: typeof import("./provider-runtime.js").buildProviderAuthDoctorHintWithPlugin;
|
||||
let buildProviderMissingAuthMessageWithPlugin: typeof import("./provider-runtime.js").buildProviderMissingAuthMessageWithPlugin;
|
||||
let buildProviderUnknownModelHintWithPlugin: typeof import("./provider-runtime.js").buildProviderUnknownModelHintWithPlugin;
|
||||
let formatProviderAuthProfileApiKeyWithPlugin: typeof import("./provider-runtime.js").formatProviderAuthProfileApiKeyWithPlugin;
|
||||
let prepareProviderExtraParams: typeof import("./provider-runtime.js").prepareProviderExtraParams;
|
||||
let resolveProviderStreamFn: typeof import("./provider-runtime.js").resolveProviderStreamFn;
|
||||
let resolveProviderCacheTtlEligibility: typeof import("./provider-runtime.js").resolveProviderCacheTtlEligibility;
|
||||
let resolveProviderBinaryThinking: typeof import("./provider-runtime.js").resolveProviderBinaryThinking;
|
||||
let resolveProviderBuiltInModelSuppression: typeof import("./provider-runtime.js").resolveProviderBuiltInModelSuppression;
|
||||
let createProviderEmbeddingProvider: typeof import("./provider-runtime.js").createProviderEmbeddingProvider;
|
||||
let resolveProviderDefaultThinkingLevel: typeof import("./provider-runtime.js").resolveProviderDefaultThinkingLevel;
|
||||
let resolveProviderModernModelRef: typeof import("./provider-runtime.js").resolveProviderModernModelRef;
|
||||
let resolveProviderSyntheticAuthWithPlugin: typeof import("./provider-runtime.js").resolveProviderSyntheticAuthWithPlugin;
|
||||
let resolveProviderUsageSnapshotWithPlugin: typeof import("./provider-runtime.js").resolveProviderUsageSnapshotWithPlugin;
|
||||
let resolveProviderCapabilitiesWithPlugin: typeof import("./provider-runtime.js").resolveProviderCapabilitiesWithPlugin;
|
||||
let resolveProviderUsageAuthWithPlugin: typeof import("./provider-runtime.js").resolveProviderUsageAuthWithPlugin;
|
||||
@@ -72,13 +76,17 @@ describe("provider-runtime", () => {
|
||||
augmentModelCatalogWithProviderPlugins,
|
||||
buildProviderAuthDoctorHintWithPlugin,
|
||||
buildProviderMissingAuthMessageWithPlugin,
|
||||
buildProviderUnknownModelHintWithPlugin,
|
||||
formatProviderAuthProfileApiKeyWithPlugin,
|
||||
prepareProviderExtraParams,
|
||||
resolveProviderStreamFn,
|
||||
resolveProviderCacheTtlEligibility,
|
||||
resolveProviderBinaryThinking,
|
||||
resolveProviderBuiltInModelSuppression,
|
||||
createProviderEmbeddingProvider,
|
||||
resolveProviderDefaultThinkingLevel,
|
||||
resolveProviderModernModelRef,
|
||||
resolveProviderSyntheticAuthWithPlugin,
|
||||
resolveProviderUsageSnapshotWithPlugin,
|
||||
resolveProviderCapabilitiesWithPlugin,
|
||||
resolveProviderUsageAuthWithPlugin,
|
||||
@@ -152,6 +160,22 @@ describe("provider-runtime", () => {
|
||||
return undefined;
|
||||
});
|
||||
const prepareDynamicModel = vi.fn(async () => undefined);
|
||||
const createStreamFn = vi.fn(() => vi.fn());
|
||||
const createEmbeddingProvider = vi.fn(async () => ({
|
||||
id: "demo",
|
||||
model: "demo-embed",
|
||||
embedQuery: async () => [1, 0, 0],
|
||||
embedBatch: async () => [[1, 0, 0]],
|
||||
client: { token: "embed-token" },
|
||||
}));
|
||||
const resolveSyntheticAuth = vi.fn(() => ({
|
||||
apiKey: "demo-local",
|
||||
source: "models.providers.demo (synthetic local key)",
|
||||
mode: "api-key" as const,
|
||||
}));
|
||||
const buildUnknownModelHint = vi.fn(
|
||||
({ modelId }: { modelId: string }) => `Use demo setup for ${modelId}`,
|
||||
);
|
||||
const prepareRuntimeAuth = vi.fn(async () => ({
|
||||
apiKey: "runtime-token",
|
||||
baseUrl: "https://runtime.example.com/v1",
|
||||
@@ -185,7 +209,13 @@ describe("provider-runtime", () => {
|
||||
...extraParams,
|
||||
transport: "auto",
|
||||
}),
|
||||
wrapStreamFn: ({ streamFn }) => streamFn,
|
||||
createStreamFn,
|
||||
wrapStreamFn: ({ streamFn, model }) => {
|
||||
expect(model).toMatchObject(MODEL);
|
||||
return streamFn;
|
||||
},
|
||||
createEmbeddingProvider,
|
||||
resolveSyntheticAuth,
|
||||
normalizeResolvedModel: ({ model }) => ({
|
||||
...model,
|
||||
api: "openai-codex-responses",
|
||||
@@ -210,6 +240,7 @@ describe("provider-runtime", () => {
|
||||
auth: [],
|
||||
buildMissingAuthMessage: () =>
|
||||
'No API key found for provider "openai". Use openai-codex/gpt-5.4.',
|
||||
buildUnknownModelHint,
|
||||
suppressBuiltInModel: ({ provider, modelId }) =>
|
||||
provider === "azure-openai-responses" && modelId === "gpt-5.3-codex-spark"
|
||||
? { suppress: true, errorMessage: "openai-codex/gpt-5.3-codex-spark" }
|
||||
@@ -270,12 +301,39 @@ describe("provider-runtime", () => {
|
||||
transport: "auto",
|
||||
});
|
||||
|
||||
expect(
|
||||
resolveProviderStreamFn({
|
||||
provider: "demo",
|
||||
context: {
|
||||
provider: "demo",
|
||||
modelId: MODEL.id,
|
||||
model: MODEL,
|
||||
},
|
||||
}),
|
||||
).toBeTypeOf("function");
|
||||
|
||||
await expect(
|
||||
createProviderEmbeddingProvider({
|
||||
provider: "demo",
|
||||
context: {
|
||||
config: {} as never,
|
||||
provider: "demo",
|
||||
model: "demo-embed",
|
||||
},
|
||||
}),
|
||||
).resolves.toMatchObject({
|
||||
id: "demo",
|
||||
model: "demo-embed",
|
||||
client: { token: "embed-token" },
|
||||
});
|
||||
|
||||
expect(
|
||||
wrapProviderStreamFn({
|
||||
provider: "demo",
|
||||
context: {
|
||||
provider: "demo",
|
||||
modelId: MODEL.id,
|
||||
model: MODEL,
|
||||
streamFn: vi.fn(),
|
||||
},
|
||||
}),
|
||||
@@ -439,12 +497,44 @@ describe("provider-runtime", () => {
|
||||
}),
|
||||
).toBe(true);
|
||||
|
||||
expect(
|
||||
resolveProviderSyntheticAuthWithPlugin({
|
||||
provider: "demo",
|
||||
context: {
|
||||
provider: "demo",
|
||||
providerConfig: {
|
||||
api: "openai-completions",
|
||||
baseUrl: "http://localhost:11434",
|
||||
models: [],
|
||||
},
|
||||
},
|
||||
}),
|
||||
).toEqual({
|
||||
apiKey: "demo-local",
|
||||
source: "models.providers.demo (synthetic local key)",
|
||||
mode: "api-key",
|
||||
});
|
||||
|
||||
expect(
|
||||
buildProviderUnknownModelHintWithPlugin({
|
||||
provider: "openai",
|
||||
env: process.env,
|
||||
context: {
|
||||
env: process.env,
|
||||
provider: "openai",
|
||||
modelId: "gpt-5.4",
|
||||
},
|
||||
}),
|
||||
).toBe("Use demo setup for gpt-5.4");
|
||||
|
||||
expectCodexMissingAuthHint(buildProviderMissingAuthMessageWithPlugin);
|
||||
expectCodexBuiltInSuppression(resolveProviderBuiltInModelSuppression);
|
||||
await expectAugmentedCodexCatalog(augmentModelCatalogWithProviderPlugins);
|
||||
|
||||
expect(prepareDynamicModel).toHaveBeenCalledTimes(1);
|
||||
expect(refreshOAuth).toHaveBeenCalledTimes(1);
|
||||
expect(resolveSyntheticAuth).toHaveBeenCalledTimes(1);
|
||||
expect(buildUnknownModelHint).toHaveBeenCalledTimes(1);
|
||||
expect(prepareRuntimeAuth).toHaveBeenCalledTimes(1);
|
||||
expect(resolveUsageAuth).toHaveBeenCalledTimes(1);
|
||||
expect(fetchUsageSnapshot).toHaveBeenCalledTimes(1);
|
||||
|
||||
@@ -15,8 +15,12 @@ import type {
|
||||
ProviderAuthDoctorHintContext,
|
||||
ProviderAugmentModelCatalogContext,
|
||||
ProviderBuildMissingAuthMessageContext,
|
||||
ProviderBuildUnknownModelHintContext,
|
||||
ProviderBuiltInModelSuppressionContext,
|
||||
ProviderCacheTtlEligibilityContext,
|
||||
ProviderCreateEmbeddingProviderContext,
|
||||
ProviderResolveSyntheticAuthContext,
|
||||
ProviderCreateStreamFnContext,
|
||||
ProviderDefaultThinkingPolicyContext,
|
||||
ProviderFetchUsageSnapshotContext,
|
||||
ProviderModernModelPolicyContext,
|
||||
@@ -234,6 +238,16 @@ export function prepareProviderExtraParams(params: {
|
||||
return resolveProviderRuntimePlugin(params)?.prepareExtraParams?.(params.context) ?? undefined;
|
||||
}
|
||||
|
||||
export function resolveProviderStreamFn(params: {
|
||||
provider: string;
|
||||
config?: OpenClawConfig;
|
||||
workspaceDir?: string;
|
||||
env?: NodeJS.ProcessEnv;
|
||||
context: ProviderCreateStreamFnContext;
|
||||
}) {
|
||||
return resolveProviderRuntimePlugin(params)?.createStreamFn?.(params.context) ?? undefined;
|
||||
}
|
||||
|
||||
export function wrapProviderStreamFn(params: {
|
||||
provider: string;
|
||||
config?: OpenClawConfig;
|
||||
@@ -244,6 +258,16 @@ export function wrapProviderStreamFn(params: {
|
||||
return resolveProviderRuntimePlugin(params)?.wrapStreamFn?.(params.context) ?? undefined;
|
||||
}
|
||||
|
||||
export async function createProviderEmbeddingProvider(params: {
|
||||
provider: string;
|
||||
config?: OpenClawConfig;
|
||||
workspaceDir?: string;
|
||||
env?: NodeJS.ProcessEnv;
|
||||
context: ProviderCreateEmbeddingProviderContext;
|
||||
}) {
|
||||
return await resolveProviderRuntimePlugin(params)?.createEmbeddingProvider?.(params.context);
|
||||
}
|
||||
|
||||
export async function prepareProviderRuntimeAuth(params: {
|
||||
provider: string;
|
||||
config?: OpenClawConfig;
|
||||
@@ -366,6 +390,26 @@ export function buildProviderMissingAuthMessageWithPlugin(params: {
|
||||
);
|
||||
}
|
||||
|
||||
export function buildProviderUnknownModelHintWithPlugin(params: {
|
||||
provider: string;
|
||||
config?: OpenClawConfig;
|
||||
workspaceDir?: string;
|
||||
env?: NodeJS.ProcessEnv;
|
||||
context: ProviderBuildUnknownModelHintContext;
|
||||
}) {
|
||||
return resolveProviderRuntimePlugin(params)?.buildUnknownModelHint?.(params.context) ?? undefined;
|
||||
}
|
||||
|
||||
export function resolveProviderSyntheticAuthWithPlugin(params: {
|
||||
provider: string;
|
||||
config?: OpenClawConfig;
|
||||
workspaceDir?: string;
|
||||
env?: NodeJS.ProcessEnv;
|
||||
context: ProviderResolveSyntheticAuthContext;
|
||||
}) {
|
||||
return resolveProviderRuntimePlugin(params)?.resolveSyntheticAuth?.(params.context) ?? undefined;
|
||||
}
|
||||
|
||||
export function resolveProviderBuiltInModelSuppression(params: {
|
||||
config?: OpenClawConfig;
|
||||
workspaceDir?: string;
|
||||
|
||||
@@ -21,7 +21,11 @@ import type {
|
||||
ChannelStructuredComponents,
|
||||
} from "../channels/plugins/types.js";
|
||||
import type { OpenClawConfig } from "../config/config.js";
|
||||
import type { CliBackendConfig, ModelProviderConfig } from "../config/types.js";
|
||||
import type {
|
||||
CliBackendConfig,
|
||||
ModelProviderAuthMode,
|
||||
ModelProviderConfig,
|
||||
} from "../config/types.js";
|
||||
import type { OperatorScope } from "../gateway/method-scopes.js";
|
||||
import type { GatewayRequestHandler } from "../gateway/server-methods/types.js";
|
||||
import type { InternalHookHandler } from "../hooks/internal-hooks.js";
|
||||
@@ -478,6 +482,22 @@ export type ProviderPrepareExtraParamsContext = {
|
||||
thinkingLevel?: ThinkLevel;
|
||||
};
|
||||
|
||||
/**
|
||||
* Provider-owned transport creation.
|
||||
*
|
||||
* Use this when the provider needs to replace pi-ai's default transport with a
|
||||
* custom StreamFn (for example a native API transport that cannot be expressed
|
||||
* as a wrapper around `streamSimple`).
|
||||
*/
|
||||
export type ProviderCreateStreamFnContext = {
|
||||
config?: OpenClawConfig;
|
||||
agentDir?: string;
|
||||
workspaceDir?: string;
|
||||
provider: string;
|
||||
modelId: string;
|
||||
model: ProviderRuntimeModel;
|
||||
};
|
||||
|
||||
/**
|
||||
* Provider-owned stream wrapper hook after OpenClaw applies its generic
|
||||
* transport-independent wrappers.
|
||||
@@ -486,9 +506,48 @@ export type ProviderPrepareExtraParamsContext = {
|
||||
* through the normal `pi-ai` stream path.
|
||||
*/
|
||||
export type ProviderWrapStreamFnContext = ProviderPrepareExtraParamsContext & {
|
||||
model?: ProviderRuntimeModel;
|
||||
streamFn?: StreamFn;
|
||||
};
|
||||
|
||||
/**
|
||||
* Generic embedding provider shape returned by provider plugins.
|
||||
*
|
||||
* Keep this aligned with the memory embedding contract without forcing the
|
||||
* plugin system to import memory internals directly.
|
||||
*/
|
||||
export type PluginEmbeddingProvider = {
|
||||
id: string;
|
||||
model: string;
|
||||
maxInputTokens?: number;
|
||||
embedQuery: (text: string) => Promise<number[]>;
|
||||
embedBatch: (texts: string[]) => Promise<number[][]>;
|
||||
embedBatchInputs?: (inputs: unknown[]) => Promise<number[][]>;
|
||||
client?: unknown;
|
||||
};
|
||||
|
||||
/**
|
||||
* Provider-owned embedding transport creation.
|
||||
*
|
||||
* Use this when a provider wants memory embeddings to live with the provider
|
||||
* plugin instead of the core memory switchboard.
|
||||
*/
|
||||
export type ProviderCreateEmbeddingProviderContext = {
|
||||
config: OpenClawConfig;
|
||||
agentDir?: string;
|
||||
workspaceDir?: string;
|
||||
provider: string;
|
||||
model: string;
|
||||
remote?: {
|
||||
baseUrl?: string;
|
||||
apiKey?: unknown;
|
||||
headers?: Record<string, string>;
|
||||
};
|
||||
providerApiKey?: string;
|
||||
outputDimensionality?: number;
|
||||
taskType?: string;
|
||||
};
|
||||
|
||||
/**
|
||||
* Provider-owned prompt-cache eligibility.
|
||||
*
|
||||
@@ -516,6 +575,22 @@ export type ProviderBuildMissingAuthMessageContext = {
|
||||
listProfileIds: (providerId: string) => string[];
|
||||
};
|
||||
|
||||
/**
|
||||
* Provider-owned unknown-model hint override.
|
||||
*
|
||||
* Runs after catalog/runtime lookup misses for the requested provider. Return a
|
||||
* hint suffix that OpenClaw should append to the generic `Unknown model`
|
||||
* error.
|
||||
*/
|
||||
export type ProviderBuildUnknownModelHintContext = {
|
||||
config?: OpenClawConfig;
|
||||
agentDir?: string;
|
||||
workspaceDir?: string;
|
||||
env: NodeJS.ProcessEnv;
|
||||
provider: string;
|
||||
modelId: string;
|
||||
};
|
||||
|
||||
/**
|
||||
* Built-in model suppression hook.
|
||||
*
|
||||
@@ -632,6 +707,17 @@ export type ProviderPluginWizardSetup = {
|
||||
initialSelections?: string[];
|
||||
message?: string;
|
||||
};
|
||||
/**
|
||||
* Optional default-model prompt policy for this auth/setup choice.
|
||||
*
|
||||
* Use this when selecting the auth choice should still force a model picker
|
||||
* even if the choice was preseeded via CLI/configure, or when "keep current"
|
||||
* would skip required provider-owned post-selection work.
|
||||
*/
|
||||
modelSelection?: {
|
||||
promptWhenAuthChoiceProvided?: boolean;
|
||||
allowKeepCurrent?: boolean;
|
||||
};
|
||||
};
|
||||
|
||||
/** Optional model-picker metadata shown in interactive provider selection flows. */
|
||||
@@ -655,6 +741,18 @@ export type ProviderModelSelectedContext = {
|
||||
workspaceDir?: string;
|
||||
};
|
||||
|
||||
export type ProviderResolveSyntheticAuthContext = {
|
||||
config?: OpenClawConfig;
|
||||
provider: string;
|
||||
providerConfig?: ModelProviderConfig;
|
||||
};
|
||||
|
||||
export type ProviderSyntheticAuthResult = {
|
||||
apiKey: string;
|
||||
source: string;
|
||||
mode: Exclude<ModelProviderAuthMode, "aws-sdk">;
|
||||
};
|
||||
|
||||
/** Text-inference provider capability registered by a plugin. */
|
||||
export type ProviderPlugin = {
|
||||
id: string;
|
||||
@@ -731,6 +829,13 @@ export type ProviderPlugin = {
|
||||
prepareExtraParams?: (
|
||||
ctx: ProviderPrepareExtraParamsContext,
|
||||
) => Record<string, unknown> | null | undefined;
|
||||
/**
|
||||
* Provider-owned transport factory.
|
||||
*
|
||||
* Use this when the provider needs a fully custom StreamFn instead of a
|
||||
* wrapper around the normal `streamSimple` path.
|
||||
*/
|
||||
createStreamFn?: (ctx: ProviderCreateStreamFnContext) => StreamFn | null | undefined;
|
||||
/**
|
||||
* Provider-owned stream wrapper applied after generic OpenClaw wrappers.
|
||||
*
|
||||
@@ -739,6 +844,19 @@ export type ProviderPlugin = {
|
||||
* transport implementation.
|
||||
*/
|
||||
wrapStreamFn?: (ctx: ProviderWrapStreamFnContext) => StreamFn | null | undefined;
|
||||
/**
|
||||
* Provider-owned embedding provider factory.
|
||||
*
|
||||
* Use this when memory embedding behavior belongs with the provider plugin
|
||||
* rather than the core embedding switchboard.
|
||||
*/
|
||||
createEmbeddingProvider?: (
|
||||
ctx: ProviderCreateEmbeddingProviderContext,
|
||||
) =>
|
||||
| Promise<PluginEmbeddingProvider | null | undefined>
|
||||
| PluginEmbeddingProvider
|
||||
| null
|
||||
| undefined;
|
||||
/**
|
||||
* Runtime auth exchange hook.
|
||||
*
|
||||
@@ -791,6 +909,14 @@ export type ProviderPlugin = {
|
||||
buildMissingAuthMessage?: (
|
||||
ctx: ProviderBuildMissingAuthMessageContext,
|
||||
) => string | null | undefined;
|
||||
/**
|
||||
* Provider-owned unknown-model hint override.
|
||||
*
|
||||
* Return a suffix when the provider wants a more specific recovery hint than
|
||||
* OpenClaw's generic `Unknown model` error after catalog/runtime lookup
|
||||
* fails.
|
||||
*/
|
||||
buildUnknownModelHint?: (ctx: ProviderBuildUnknownModelHintContext) => string | null | undefined;
|
||||
/**
|
||||
* Provider-owned built-in model suppression.
|
||||
*
|
||||
@@ -882,6 +1008,16 @@ export type ProviderPlugin = {
|
||||
buildAuthDoctorHint?: (
|
||||
ctx: ProviderAuthDoctorHintContext,
|
||||
) => string | Promise<string | null | undefined> | null | undefined;
|
||||
/**
|
||||
* Provider-owned synthetic auth marker.
|
||||
*
|
||||
* Use this when the provider can operate without a real secret for certain
|
||||
* configured local/self-hosted cases and wants auth resolution to treat that
|
||||
* config as available.
|
||||
*/
|
||||
resolveSyntheticAuth?: (
|
||||
ctx: ProviderResolveSyntheticAuthContext,
|
||||
) => ProviderSyntheticAuthResult | null | undefined;
|
||||
onModelSelected?: (ctx: ProviderModelSelectedContext) => Promise<void>;
|
||||
};
|
||||
|
||||
|
||||
Reference in New Issue
Block a user