Files
pi-commandcode-provider/index.ts
T
Patrick Wozniak 59322e4ffa perf(models): start from the cached catalog and refresh in the background
Every host start awaited a full catalog request before the provider
registered, costing one HTTPS round-trip (up to the discovery timeout on a
hanging connection) even when a cache written seconds earlier was on disk.

Register the cached catalog immediately and run the live refresh in the
background; the live result re-registers the provider when it arrives. A
first start without a cache still awaits the live catalog. The background
refresh is aborted on session_shutdown so print mode does not wait for it.

Closes #63

(cherry picked from commit 142191f9420fe90460cdc3cbae70e26d0c000860)
2026-09-01 23:29:33 +02:00

181 lines
6.0 KiB
TypeScript

/**
* Command Code provider for pi.
*
* Uses Command Code's documented Provider API:
* https://api.commandcode.ai/provider/v1
*/
import { AssistantMessageEventStream } from "@earendil-works/pi-ai"
import {
registerApiProvider,
streamSimple as streamNativeProvider,
type ApiStreamSimpleFunction,
} from "@earendil-works/pi-ai/compat"
import {
getAgentDir,
type ExtensionAPI,
type ExtensionCommandContext,
type ProviderConfig,
} from "@earendil-works/pi-coding-agent"
import { join } from "node:path"
import { getConfiguredApiKey } from "./src/api-key.ts"
import { createStreamCommandCode } from "./src/core.ts"
import { calculateCommandCodeCost } from "./src/cost.ts"
import {
apiForModelId,
baseUrlForModel,
DEFAULT_MODELS_URL,
DEFAULT_PROVIDER_API_BASE,
getModelsTimeoutMs,
inputModalitiesForModel,
loadCachedCommandCodeModels,
loadCommandCodeModels,
MODEL_EFFORTS,
thinkingMetadataForModel,
type CommandCodeModel,
} from "./src/models.ts"
import { getApiKey as getOAuthApiKey, login, refreshToken } from "./src/oauth.ts"
import { normalizeCommandCodeMessage } from "./src/overflow.ts"
import { MODEL_COSTS, ZERO_MODEL_COST } from "./src/pricing.ts"
import { registerCommandCodeQuota } from "./src/quota-command.ts"
import { createCommandCodeRuntime } from "./src/runtime.ts"
import { createCommandCodeTransportRouter } from "./src/transport.ts"
const COMMAND_CODE_API = "commandcode-custom"
const COMPAT_SOURCE_ID = "pi-commandcode-provider"
function commandCodeHeaders(): Record<string, string> | undefined {
if (process.env.CMD_ZDR === "1" || process.env.COMMANDCODE_ZDR === "1") {
return { "x-cmd-zdr": "1" }
}
return undefined
}
function createProviderConfig(
models: readonly CommandCodeModel[],
apiBase: string,
streamCommandCode: ProviderConfig["streamSimple"],
): ProviderConfig {
const headers = commandCodeHeaders()
return {
name: "Command Code",
baseUrl: apiBase,
apiKey: getConfiguredApiKey() ?? "$COMMAND_CODE_API_KEY",
api: COMMAND_CODE_API,
streamSimple: streamCommandCode,
headers,
oauth: {
name: "Command Code",
login,
refreshToken,
getApiKey: getOAuthApiKey,
},
models: models.map((model) => ({
id: model.id,
name: model.name,
api: COMMAND_CODE_API,
baseUrl: baseUrlForModel(apiBase, model.api),
reasoning: model.reasoning,
...(thinkingMetadataForModel(model.id) ?? {}),
input: [...inputModalitiesForModel(model.id)],
cost: MODEL_COSTS[model.id] ?? ZERO_MODEL_COST,
contextWindow: model.contextWindow,
maxTokens: model.maxTokens,
headers,
compat:
model.api === "openai-completions"
? {
supportsStore: false,
supportsDeveloperRole: false,
supportsReasoningEffort: MODEL_EFFORTS[model.id] !== undefined,
maxTokensField: "max_tokens",
}
: {
supportsEagerToolInputStreaming: false,
supportsLongCacheRetention: false,
supportsCacheControlOnTools: false,
supportsToolReferences: false,
...(model.reasoning ? { forceAdaptiveThinking: true } : {}),
},
})),
}
}
function legacyApiBase(providerApiBase: string): string {
return providerApiBase.replace(/\/provider\/v1\/?$/, "")
}
export default async function (pi: ExtensionAPI) {
const apiBase = process.env.COMMANDCODE_API_BASE ?? DEFAULT_PROVIDER_API_BASE
const modelsUrl = process.env.COMMANDCODE_MODELS_URL ?? DEFAULT_MODELS_URL
const modelsTimeoutMs = getModelsTimeoutMs()
const modelsCachePath =
process.env.COMMANDCODE_MODELS_CACHE ?? join(getAgentDir(), "commandcode-models.json")
const streamGenerate = createStreamCommandCode({
createStream: () => new AssistantMessageEventStream(),
calculateCost: calculateCommandCodeCost,
apiBase: legacyApiBase(apiBase),
})
const transport = createCommandCodeTransportRouter({
createStream: () => new AssistantMessageEventStream(),
streamProvider: (model, context, options) =>
streamNativeProvider(
{ ...model, api: apiForModelId(model.id), compat: model.compatConfig ?? model.compat },
context,
options,
),
streamGenerate,
})
// pi dispatches the main chat through the registered provider, but sibling
// extensions that call `streamSimple` from `@earendil-works/pi-ai/compat`
// with a Command Code model resolve `model.api` through the compat
// api-registry, which knows nothing about extension providers. Register the
// custom api there so those calls reach the same transport. The registry
// resolves no credentials for extension providers, so fall back to the
// configured key when the caller passes none.
const compatStream: ApiStreamSimpleFunction = (model, context, options) =>
transport.stream(
model,
context,
options?.apiKey ? options : { ...options, apiKey: getConfiguredApiKey() },
) as AssistantMessageEventStream
registerApiProvider(
{ api: COMMAND_CODE_API, stream: compatStream, streamSimple: compatStream },
COMPAT_SOURCE_ID,
)
pi.on("message_end", async (event, ctx) => {
if (event.message.role !== "assistant") return
const normalized = normalizeCommandCodeMessage(event.message, ctx.model?.provider)
return normalized ? { message: normalized.message } : undefined
})
registerCommandCodeQuota(pi, {
apiBase: legacyApiBase(apiBase),
headers: commandCodeHeaders(),
})
const runtime = createCommandCodeRuntime<ProviderConfig, ExtensionCommandContext>(pi, {
endpoint: modelsUrl,
cachePath: modelsCachePath,
loadModels: (signal) =>
loadCommandCodeModels({
url: modelsUrl,
cachePath: modelsCachePath,
timeoutMs: modelsTimeoutMs,
signal,
}),
loadCachedModels: () => loadCachedCommandCodeModels(modelsCachePath),
createProviderConfig: (models) => createProviderConfig(models, apiBase, transport.stream),
getTransport: transport.getTransport,
})
pi.on("session_shutdown", () => {
runtime.dispose()
})
await runtime.initialize()
}