a7d73ebb62
Built-in design generation now runs as an agentic MCP tool-loop (reusing the agent-rs BuiltInProvider), gated behind OPENPENCIL_DESIGN_AGENT_LOOP / the Settings experimental toggle; the orchestrator stays the default. - design-agent system prompt + in-process design toolset (parity-locked with the MCP surface) + flag-gated Intent::Design routing - spawn_agents execution as sequential sub-loops + live creation-mode badges (per-agent glow + 'N/M designing...' header) - new MCP tools: get_guidelines, ToolSearch, get_screenshot, get_editor_state, export_nodes, spawn_agents; style-guide local audit - #27 AI panel restyle: rounded tool cards + green check-rings, gray user bubbles, model-pill bottom toolbar, header, empty-state pills, the PARALLEL AGENTS (agent_team_size) 1x-6x chip dropdown - multi-chat tabs: ChatSessions model (Deref-to-active) + tab row UI (switch / close / + / Cmd+T) with each run bound to its tab Large checkpoint commit spanning the working tree (Rust shell crates).
60 lines
2.2 KiB
TypeScript
60 lines
2.2 KiB
TypeScript
/**
|
|
* 方舟 (Volcengine Ark) Coding Plan wrapper. One API key covers
|
|
* multiple third-party models hosted on Ark's coding tier —
|
|
* currently GLM-5.1 and KIMI-K2.6. Cribbed from builtin-provider-
|
|
* presets.ts `ark-coding` preset.
|
|
*
|
|
* Default baseURL: https://ark.cn-beijing.volces.com/api/coding/v3
|
|
* Override via `ARK_BASE_URL` if Volcengine ever ships a regional
|
|
* variant (none published as of 2026-04-22).
|
|
*
|
|
* Key env var: `ARK_CODING_KEY` — Volcengine's UUID-format ARK
|
|
* access key. NOT committed to the repo; export in your shell
|
|
* before running `bun scripts/ab-corpus/run.ts --live`.
|
|
*/
|
|
|
|
import type { ChatCallResult } from './openai-compat';
|
|
import { callOpenAICompat } from './openai-compat';
|
|
|
|
const ARK_BASE_URL = process.env.ARK_BASE_URL ?? 'https://ark.cn-beijing.volces.com/api/coding/v3';
|
|
const ARK_API_KEY_ENV = 'ARK_CODING_KEY';
|
|
|
|
export interface CallArkArgs {
|
|
model: string;
|
|
system: string;
|
|
user: string;
|
|
temperature?: number;
|
|
maxTokens?: number;
|
|
}
|
|
|
|
export async function callArk(args: CallArkArgs): Promise<ChatCallResult> {
|
|
const apiKey = process.env[ARK_API_KEY_ENV];
|
|
if (!apiKey) {
|
|
throw new Error(
|
|
`${ARK_API_KEY_ENV} not set — export it (Volcengine 方舟 CP UUID key) before running with --live`,
|
|
);
|
|
}
|
|
// Kimi-K2.6 on Ark CP is markedly more brownout-prone than GLM-5.1
|
|
// — ab-v5 (2026-05-02) cut its T arm garbage rate to 17% from
|
|
// ab-v2's 42%, but it still contributed 9/14 of the T arm garbage
|
|
// and ARK's empty-content / 120s-timeout were the only failure
|
|
// modes. Bump the kimi branch to retries=3 (extra 4000ms backoff
|
|
// attempt) + timeoutMs=180000 (60s headroom) so slow-but-eventually-
|
|
// OK responses don't get clipped. GLM-5.1 keeps the existing
|
|
// retries=2 + 120s defaults — its garbage rate is already <2%,
|
|
// pushing those would just burn budget on healthy calls.
|
|
const isKimi = /^kimi/i.test(args.model);
|
|
return callOpenAICompat({
|
|
baseURL: ARK_BASE_URL,
|
|
apiKey,
|
|
model: args.model,
|
|
system: args.system,
|
|
user: args.user,
|
|
temperature: args.temperature,
|
|
maxTokens: args.maxTokens,
|
|
label: 'ark',
|
|
retries: isKimi ? 3 : 2,
|
|
timeoutMs: isKimi ? 180_000 : undefined,
|
|
});
|
|
}
|