refactor(pi-devin): switch from npm+patch to vendored local fork
- Vendor the patched pi-devin@0.1.2 tree at apps/pi-devin/upstream/ (third-party archive -> upstream_fork); git history replaces the patch layer as the change archive. Includes the devin-desktop identity fix and normalized catalog variant mapping from the previous patch. - build-pi-package.sh now verifies source.sha256 against host-local-packages.json and npm-packs the vendored tree; drop the npm fetch + patch stage. - install-pi-package.sh strips the npm-pack package/ prefix to match the playbook's unarchive behavior. - Deployed to a100 via deploy_pi_devin_package (--limit a100, R2 then R3); remote verified to carry the fork source.
This commit is contained in:
Shujakuinkuraudo
+307
-178
@@ -4,15 +4,15 @@ import {
|
||||
type Api,
|
||||
type AssistantMessage,
|
||||
type AssistantMessageEventStream,
|
||||
type Context,
|
||||
type Model,
|
||||
type SimpleStreamOptions,
|
||||
type TranscriptContext,
|
||||
calculateCost,
|
||||
createAssistantMessageEventStream,
|
||||
parseStreamingJson,
|
||||
} from "@earendil-works/pi-ai";
|
||||
import { mapContextToChat, type ChatHistoryItem, type ContentPart, type ToolDef } from "./context-map.js";
|
||||
import { getCachedUserJwt } from "./jwt.js";
|
||||
import { clampMaxTokensToContext } from "./context-estimate.js";
|
||||
import { mapContextToChat, type ChatHistoryItem, type ToolDef } from "./context-map.js";
|
||||
import { anySignal, getCachedUserJwt } from "./jwt.js";
|
||||
import { buildMetadata } from "./metadata.js";
|
||||
import { resolveModelUid } from "./models.js";
|
||||
import {
|
||||
@@ -24,32 +24,40 @@ import {
|
||||
iterFields,
|
||||
} from "./wire.js";
|
||||
|
||||
const SOURCE_BY_ROLE: Record<ChatHistoryItem["role"], number> = {
|
||||
const SOURCE_BY_ROLE: Record<string, number> = {
|
||||
user: 1,
|
||||
assistant: 2,
|
||||
tool: 4,
|
||||
};
|
||||
|
||||
/** Connect frame payloads larger than this indicate a corrupt length prefix. */
|
||||
const MAX_CONNECT_FRAME_PAYLOAD = 16 * 1024 * 1024;
|
||||
/**
|
||||
* Recovery heuristic for opaque Devin `invalid_argument`/`internal error`
|
||||
* trailers: compactable message history this large is likely the cause, so the
|
||||
* failure is reported in a shape Pi recognizes as context overflow.
|
||||
*/
|
||||
const LARGE_HISTORY_RECOVERY_BYTES = 512 * 1024;
|
||||
|
||||
export type CloudChatEvent =
|
||||
| { kind: "text"; text: string }
|
||||
| { kind: "reasoning"; text: string }
|
||||
| { kind: "signature"; text: string }
|
||||
| { kind: "thinking_redacted" }
|
||||
| { kind: "message_id"; id: string }
|
||||
| { kind: "actual_model"; uid: string }
|
||||
| { kind: "tool_call_start"; id: string; name: string }
|
||||
| { kind: "tool_call_args"; argsDelta: string; id?: string }
|
||||
| { kind: "finish"; reason: "stop" | "tool_calls" | "length" | "content_filter" }
|
||||
| { kind: "finish"; reason: "stop" | "tool_calls" | "length" | "content_filter" | "error" }
|
||||
| {
|
||||
kind: "usage";
|
||||
promptTokens?: number;
|
||||
completionTokens?: number;
|
||||
totalTokens?: number;
|
||||
cachedInputTokens?: number;
|
||||
cacheCreationInputTokens?: number;
|
||||
input?: number;
|
||||
output?: number;
|
||||
cacheRead?: number;
|
||||
cacheWrite?: number;
|
||||
modelUid?: string;
|
||||
};
|
||||
|
||||
function normalizeContent(content: string | ContentPart[]): ContentPart[] {
|
||||
if (typeof content === "string") return [{ type: "text", text: content }];
|
||||
return content;
|
||||
}
|
||||
|
||||
function encodeImageData(img: { mimeType?: string; base64Data?: string }): Buffer {
|
||||
return Buffer.concat([
|
||||
encodeString(1, img.base64Data ?? ""),
|
||||
@@ -61,39 +69,41 @@ function encodeChatToolCall(tc: { id: string; name: string; arguments: string })
|
||||
return Buffer.concat([encodeString(1, tc.id), encodeString(2, tc.name), encodeString(3, tc.arguments)]);
|
||||
}
|
||||
|
||||
function encodeChatMessagePrompt(
|
||||
content: ContentPart[],
|
||||
source: number,
|
||||
opts?: { toolCallId?: string; toolCalls?: Array<{ id: string; name: string; arguments: string }> },
|
||||
): Buffer {
|
||||
const text = content
|
||||
export function encodeChatMessagePrompt(item: ChatHistoryItem, messageId: string): Buffer {
|
||||
const text = item.content
|
||||
.filter((part) => part.type === "text")
|
||||
.map((part) => part.text ?? "")
|
||||
.join("\n");
|
||||
const parts: Buffer[] = [
|
||||
encodeVarintField(2, source),
|
||||
encodeString(1, messageId),
|
||||
encodeVarintField(2, SOURCE_BY_ROLE[item.role] ?? 1),
|
||||
encodeString(3, text),
|
||||
encodeVarintField(4, Math.max(1, Math.floor(text.length / 4))),
|
||||
encodeVarintField(5, 1),
|
||||
];
|
||||
if (opts?.toolCallId) parts.push(encodeString(7, opts.toolCallId));
|
||||
for (const tc of opts?.toolCalls ?? []) parts.push(encodeMessage(6, encodeChatToolCall(tc)));
|
||||
for (const img of content.filter((part) => part.type === "image")) {
|
||||
for (const tc of item.tool_calls ?? []) parts.push(encodeMessage(6, encodeChatToolCall(tc)));
|
||||
if (item.tool_call_id) parts.push(encodeString(7, item.tool_call_id));
|
||||
if (item.tool_result_is_error) parts.push(encodeVarintField(9, 1));
|
||||
for (const img of item.content.filter((part) => part.type === "image")) {
|
||||
parts.push(encodeMessage(10, encodeImageData(img)));
|
||||
}
|
||||
if (item.thinking) parts.push(encodeString(11, item.thinking));
|
||||
if (item.signature) parts.push(encodeString(12, item.signature));
|
||||
if (item.thinking_redacted) parts.push(encodeVarintField(13, 1));
|
||||
return Buffer.concat(parts);
|
||||
}
|
||||
|
||||
function encodeCompletionConfiguration(maxOutputTokens?: number): Buffer {
|
||||
function encodeCompletionConfiguration(args: {
|
||||
maxTokens: number;
|
||||
temperature: number;
|
||||
topP: number;
|
||||
topK: number;
|
||||
}): Buffer {
|
||||
return Buffer.concat([
|
||||
encodeVarintField(1, 1),
|
||||
encodeVarintField(2, 64_000),
|
||||
encodeVarintField(3, maxOutputTokens ?? 128_000),
|
||||
encodeFixed64Field(5, 0.7),
|
||||
encodeFixed64Field(6, 0.95),
|
||||
encodeVarintField(7, 50),
|
||||
encodeFixed64Field(8, 1.0),
|
||||
encodeFixed64Field(11, 1.0),
|
||||
encodeVarintField(2, args.maxTokens),
|
||||
encodeVarintField(3, 400),
|
||||
encodeFixed64Field(5, args.temperature),
|
||||
encodeVarintField(7, args.topK),
|
||||
encodeFixed64Field(8, args.topP),
|
||||
]);
|
||||
}
|
||||
|
||||
@@ -106,7 +116,7 @@ function encodeToolDef(tool: ToolDef): Buffer {
|
||||
]);
|
||||
}
|
||||
|
||||
function buildGetChatMessageRequest(args: {
|
||||
export interface WireRequestArgs {
|
||||
apiKey: string;
|
||||
userJwt: string;
|
||||
modelUid: string;
|
||||
@@ -118,8 +128,13 @@ function buildGetChatMessageRequest(args: {
|
||||
sessionId: string;
|
||||
requestId: bigint;
|
||||
triggerId: string;
|
||||
maxOutputTokens?: number;
|
||||
}): Buffer {
|
||||
maxTokens: number;
|
||||
temperature: number;
|
||||
topP: number;
|
||||
topK: number;
|
||||
}
|
||||
|
||||
export function buildGetChatMessageRequest(args: WireRequestArgs): { body: Buffer; historyBytes: number } {
|
||||
const metadata = buildMetadata({
|
||||
apiKey: args.apiKey,
|
||||
userJwt: args.userJwt,
|
||||
@@ -127,36 +142,46 @@ function buildGetChatMessageRequest(args: {
|
||||
requestId: args.requestId,
|
||||
triggerId: args.triggerId,
|
||||
});
|
||||
const prompts = args.messages.map((message) =>
|
||||
encodeMessage(
|
||||
3,
|
||||
encodeChatMessagePrompt(normalizeContent(message.content), SOURCE_BY_ROLE[message.role], {
|
||||
toolCallId: message.role === "tool" ? message.tool_call_id : undefined,
|
||||
toolCalls: message.role === "assistant" ? message.tool_calls : undefined,
|
||||
}),
|
||||
),
|
||||
);
|
||||
return Buffer.concat([
|
||||
const promptFields: Buffer[] = [];
|
||||
args.messages.forEach((message, index) => {
|
||||
const messageId = message.message_id ?? `${args.cascadeId}-${index}`;
|
||||
promptFields.push(encodeMessage(3, encodeChatMessagePrompt(message, messageId)));
|
||||
});
|
||||
const history = Buffer.concat(promptFields);
|
||||
const body = Buffer.concat([
|
||||
encodeMessage(1, metadata),
|
||||
...(args.systemPrompt ? [encodeString(2, args.systemPrompt)] : []),
|
||||
...prompts,
|
||||
args.systemPrompt ? encodeString(2, args.systemPrompt) : Buffer.alloc(0),
|
||||
history,
|
||||
encodeVarintField(7, 5),
|
||||
encodeMessage(8, encodeCompletionConfiguration(args.maxOutputTokens)),
|
||||
encodeMessage(8, encodeCompletionConfiguration(args)),
|
||||
...(args.tools ?? []).map((tool) => encodeMessage(10, encodeToolDef(tool))),
|
||||
encodeString(16, args.cascadeId),
|
||||
encodeString(17, args.promptId),
|
||||
encodeVarintField(20, 1),
|
||||
encodeString(21, args.modelUid),
|
||||
encodeString(22, args.promptId),
|
||||
]);
|
||||
return { body, historyBytes: history.length };
|
||||
}
|
||||
|
||||
function* decodeChatFrame(proto: Buffer): Generator<CloudChatEvent> {
|
||||
export function* decodeChatFrame(proto: Buffer): Generator<CloudChatEvent> {
|
||||
for (const field of iterFields(proto)) {
|
||||
if (field.num === 3 && field.wire === 2 && Buffer.isBuffer(field.value)) {
|
||||
if (field.num === 1 && field.wire === 2 && Buffer.isBuffer(field.value)) {
|
||||
const id = field.value.toString("utf8");
|
||||
if (id) yield { kind: "message_id", id };
|
||||
} else if (field.num === 3 && field.wire === 2 && Buffer.isBuffer(field.value)) {
|
||||
const text = field.value.toString("utf8");
|
||||
if (text) yield { kind: "text", text };
|
||||
} else if (field.num === 9 && field.wire === 2 && Buffer.isBuffer(field.value)) {
|
||||
const text = field.value.toString("utf8");
|
||||
if (text) yield { kind: "reasoning", text };
|
||||
} else if (field.num === 10 && field.wire === 2 && Buffer.isBuffer(field.value)) {
|
||||
const signature = field.value.toString("utf8");
|
||||
if (signature) yield { kind: "signature", text: signature };
|
||||
} else if (field.num === 11 && field.wire === 0 && field.value) {
|
||||
yield { kind: "thinking_redacted" };
|
||||
} else if (field.num === 23 && field.wire === 2 && Buffer.isBuffer(field.value)) {
|
||||
const uid = field.value.toString("utf8");
|
||||
if (uid) yield { kind: "actual_model", uid };
|
||||
} else if (field.num === 6 && field.wire === 2 && Buffer.isBuffer(field.value)) {
|
||||
let id: string | undefined;
|
||||
let name: string | undefined;
|
||||
@@ -173,59 +198,77 @@ function* decodeChatFrame(proto: Buffer): Generator<CloudChatEvent> {
|
||||
if (argsDelta !== undefined) yield { kind: "tool_call_args", argsDelta, ...(id ? { id } : {}) };
|
||||
} else if (field.num === 5 && field.wire === 0) {
|
||||
const value = Number(field.value);
|
||||
let reason: Extract<CloudChatEvent, { kind: "finish" }>['reason'] = "stop";
|
||||
let reason: Extract<CloudChatEvent, { kind: "finish" }>["reason"] = "stop";
|
||||
if (value === 10) reason = "tool_calls";
|
||||
else if (value === 1 || value === 3 || value === 5) reason = "length";
|
||||
else if (value === 11) reason = "content_filter";
|
||||
else if (value === 1 || value === 3) reason = "length";
|
||||
else if (value === 12 || value === 13) reason = "error";
|
||||
yield { kind: "finish", reason };
|
||||
} else if (field.num === 28 && field.wire === 2 && Buffer.isBuffer(field.value)) {
|
||||
const usage = decodeUsage(field.value);
|
||||
} else if (field.num === 7 && field.wire === 2 && Buffer.isBuffer(field.value)) {
|
||||
const usage = decodeUsageStats(field.value);
|
||||
if (usage) yield usage;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
function decodeUsage(buf: Buffer): CloudChatEvent | null {
|
||||
let promptTokens: number | undefined;
|
||||
let completionTokens: number | undefined;
|
||||
let cachedInputTokens: number | undefined;
|
||||
let cacheCreationInputTokens: number | undefined;
|
||||
function decodeUsageStats(buf: Buffer): CloudChatEvent | null {
|
||||
let input: number | undefined;
|
||||
let output: number | undefined;
|
||||
let cacheWrite: number | undefined;
|
||||
let cacheRead: number | undefined;
|
||||
let modelUid: string | undefined;
|
||||
for (const field of iterFields(buf)) {
|
||||
if (field.num !== 2 || field.wire !== 2 || !Buffer.isBuffer(field.value)) continue;
|
||||
let metric: string | undefined;
|
||||
let value: number | undefined;
|
||||
for (const inner of iterFields(field.value)) {
|
||||
if (inner.num === 5 && inner.wire === 2 && Buffer.isBuffer(inner.value)) {
|
||||
metric = inner.value.toString("utf8");
|
||||
} else if (inner.num === 4 && inner.wire === 2 && Buffer.isBuffer(inner.value)) {
|
||||
for (const dim of iterFields(inner.value)) {
|
||||
if (dim.num === 2 && dim.wire === 5 && Buffer.isBuffer(dim.value)) {
|
||||
value = dim.value.readFloatLE(0);
|
||||
}
|
||||
}
|
||||
}
|
||||
if (field.wire === 0) {
|
||||
const n = Number(field.value);
|
||||
if (field.num === 2) input = n;
|
||||
else if (field.num === 3) output = n;
|
||||
else if (field.num === 4) cacheWrite = n;
|
||||
else if (field.num === 5) cacheRead = n;
|
||||
} else if (field.num === 9 && field.wire === 2 && Buffer.isBuffer(field.value)) {
|
||||
modelUid = field.value.toString("utf8");
|
||||
}
|
||||
if (!metric || value === undefined || !Number.isFinite(value)) continue;
|
||||
const n = Math.round(value);
|
||||
if (metric === "input_tokens") promptTokens = n;
|
||||
else if (metric === "output_tokens") completionTokens = n;
|
||||
else if (metric.includes("cached") || metric.includes("cache_read")) cachedInputTokens = n;
|
||||
else if (metric.includes("cache_creation")) cacheCreationInputTokens = n;
|
||||
}
|
||||
if ([promptTokens, completionTokens, cachedInputTokens, cacheCreationInputTokens].every((n) => n === undefined)) return null;
|
||||
return {
|
||||
kind: "usage",
|
||||
promptTokens,
|
||||
completionTokens,
|
||||
totalTokens: (promptTokens ?? 0) + (completionTokens ?? 0) + (cachedInputTokens ?? 0) + (cacheCreationInputTokens ?? 0),
|
||||
cachedInputTokens,
|
||||
cacheCreationInputTokens,
|
||||
};
|
||||
if (input === undefined && output === undefined && cacheRead === undefined && cacheWrite === undefined) {
|
||||
return null;
|
||||
}
|
||||
return { kind: "usage", input, output, cacheRead, cacheWrite, modelUid };
|
||||
}
|
||||
|
||||
export interface TrailerError {
|
||||
code: string;
|
||||
message: string;
|
||||
}
|
||||
|
||||
export function readTrailerError(text: string): TrailerError | null {
|
||||
if (!text.includes('"error"')) return null;
|
||||
try {
|
||||
const parsed = JSON.parse(text) as { error?: { code?: unknown; message?: unknown } };
|
||||
if (!parsed.error || typeof parsed.error !== "object") return null;
|
||||
return {
|
||||
code: typeof parsed.error.code === "string" ? parsed.error.code : "",
|
||||
message: typeof parsed.error.message === "string" ? parsed.error.message : "",
|
||||
};
|
||||
} catch {
|
||||
return { code: "", message: text.slice(0, 512) };
|
||||
}
|
||||
}
|
||||
|
||||
const sessionCache = new Map<string, { sessionId: string; cascadeId: string }>();
|
||||
|
||||
function sessionIds(apiKey: string, host: string) {
|
||||
export function sessionIds(
|
||||
apiKey: string,
|
||||
host: string,
|
||||
options?: { sessionId?: string; cacheRetention?: string },
|
||||
): { sessionId: string; cascadeId: string } {
|
||||
// Compaction and other no-cache callers must not share or reuse a cascade.
|
||||
if (options?.cacheRetention === "none") {
|
||||
return { sessionId: randomUUID(), cascadeId: randomUUID() };
|
||||
}
|
||||
// Pi threads its session id through so each session tree keeps its own
|
||||
// Devin cascade instead of sharing one process-wide cascade.
|
||||
if (options?.sessionId) {
|
||||
return { sessionId: options.sessionId, cascadeId: options.sessionId };
|
||||
}
|
||||
const key = `${host}\x1f${apiKey}`;
|
||||
let ids = sessionCache.get(key);
|
||||
if (!ids) {
|
||||
@@ -235,45 +278,41 @@ function sessionIds(apiKey: string, host: string) {
|
||||
return ids;
|
||||
}
|
||||
|
||||
function mergeHeaders(extra: Record<string, string | null> | undefined): Record<string, string> {
|
||||
const headers = new Map<string, string>([
|
||||
["content-type", "application/connect+proto"],
|
||||
["connect-protocol-version", "1"],
|
||||
["connect-accept-encoding", "gzip"],
|
||||
["user-agent", "connect-go/1.18.1 (go1.26.3)"],
|
||||
]);
|
||||
for (const [name, value] of Object.entries(extra ?? {})) {
|
||||
if (value == null) headers.delete(name.toLowerCase());
|
||||
else headers.set(name.toLowerCase(), value);
|
||||
}
|
||||
return Object.fromEntries(headers);
|
||||
}
|
||||
|
||||
async function* streamChatEvents(args: {
|
||||
apiKey: string;
|
||||
host: string;
|
||||
modelUid: string;
|
||||
systemPrompt?: string;
|
||||
messages: ChatHistoryItem[];
|
||||
tools?: ToolDef[];
|
||||
maxOutputTokens?: number;
|
||||
request: { body: Buffer; historyBytes: number };
|
||||
headers?: Record<string, string | null>;
|
||||
signal?: AbortSignal;
|
||||
fetchFn?: typeof fetch;
|
||||
onResponse?: (response: { status: number; headers: Record<string, string> }) => void | Promise<void>;
|
||||
}): AsyncGenerator<CloudChatEvent> {
|
||||
const host = args.host.replace(/\/$/, "");
|
||||
const userJwt = await getCachedUserJwt(args.apiKey, host, args.signal);
|
||||
const ids = sessionIds(args.apiKey, host);
|
||||
const proto = buildGetChatMessageRequest({
|
||||
apiKey: args.apiKey,
|
||||
userJwt,
|
||||
modelUid: args.modelUid,
|
||||
systemPrompt: args.systemPrompt,
|
||||
messages: args.messages,
|
||||
tools: args.tools,
|
||||
cascadeId: ids.cascadeId,
|
||||
promptId: randomUUID(),
|
||||
sessionId: ids.sessionId,
|
||||
requestId: BigInt(Date.now()),
|
||||
triggerId: randomUUID(),
|
||||
maxOutputTokens: args.maxOutputTokens,
|
||||
});
|
||||
|
||||
const resp = await fetch(`${host}/exa.api_server_pb.ApiServerService/GetChatMessage`, {
|
||||
method: "POST",
|
||||
headers: {
|
||||
"Content-Type": "application/connect+proto",
|
||||
"Connect-Protocol-Version": "1",
|
||||
"Connect-Content-Encoding": "gzip",
|
||||
"Connect-Accept-Encoding": "gzip",
|
||||
const resp = await (args.fetchFn ?? fetch)(
|
||||
`${host}/exa.api_server_pb.ApiServerService/GetChatMessage`,
|
||||
{
|
||||
method: "POST",
|
||||
headers: mergeHeaders(args.headers),
|
||||
body: new Uint8Array(frameConnectStream(args.request.body, false)),
|
||||
signal: args.signal,
|
||||
},
|
||||
body: new Uint8Array(frameConnectStream(proto, true)),
|
||||
signal: args.signal,
|
||||
});
|
||||
);
|
||||
if (args.onResponse) {
|
||||
await args.onResponse({ status: resp.status, headers: Object.fromEntries(resp.headers.entries()) });
|
||||
}
|
||||
if (!resp.ok) {
|
||||
throw new Error(`GetChatMessage HTTP ${resp.status}: ${(await resp.text()).slice(0, 300)}`);
|
||||
}
|
||||
@@ -283,7 +322,8 @@ async function* streamChatEvents(args: {
|
||||
const queue: Buffer[] = [];
|
||||
let queued = 0;
|
||||
let sawEos = false;
|
||||
let trailerError: string | null = null;
|
||||
let producedOutput = false;
|
||||
let trailerError: TrailerError | null = null;
|
||||
|
||||
const peek = (n: number): Buffer | null => {
|
||||
if (queued < n) return null;
|
||||
@@ -331,6 +371,9 @@ async function* streamChatEvents(args: {
|
||||
if (!header) break;
|
||||
const flags = header[0];
|
||||
const len = header.readUInt32BE(1);
|
||||
if (len > MAX_CONNECT_FRAME_PAYLOAD) {
|
||||
throw new Error(`Devin stream frame too large: ${len} bytes`);
|
||||
}
|
||||
if (queued < 5 + len) break;
|
||||
drop(5);
|
||||
const raw = peek(len) ?? Buffer.alloc(0);
|
||||
@@ -339,18 +382,15 @@ async function* streamChatEvents(args: {
|
||||
if (flags & 0x01) payload = zlib.gunzipSync(raw);
|
||||
if (flags & 0x02) {
|
||||
sawEos = true;
|
||||
const text = payload.toString("utf8");
|
||||
if (text.includes('"error"')) {
|
||||
try {
|
||||
const parsed = JSON.parse(text) as { error?: { message?: string } };
|
||||
trailerError = parsed.error?.message ?? text;
|
||||
} catch {
|
||||
trailerError = text;
|
||||
}
|
||||
}
|
||||
trailerError = readTrailerError(payload.toString("utf8").trim());
|
||||
continue;
|
||||
}
|
||||
yield* decodeChatFrame(payload);
|
||||
for (const event of decodeChatFrame(payload)) {
|
||||
if (event.kind === "text" || event.kind === "reasoning" || event.kind === "tool_call_start") {
|
||||
producedOutput = true;
|
||||
}
|
||||
yield event;
|
||||
}
|
||||
}
|
||||
}
|
||||
} finally {
|
||||
@@ -360,19 +400,35 @@ async function* streamChatEvents(args: {
|
||||
// ignore
|
||||
}
|
||||
try {
|
||||
await resp.body?.cancel();
|
||||
void resp.body?.cancel();
|
||||
} catch {
|
||||
// ignore
|
||||
}
|
||||
}
|
||||
|
||||
if (trailerError) throw new Error(trailerError);
|
||||
if (trailerError) {
|
||||
if (
|
||||
!producedOutput &&
|
||||
trailerError.code.toLowerCase() === "invalid_argument" &&
|
||||
/\binternal error\b/i.test(trailerError.message) &&
|
||||
args.request.historyBytes >= LARGE_HISTORY_RECOVERY_BYTES
|
||||
) {
|
||||
throw new Error(
|
||||
`Devin stream rejected: ${trailerError.message} (message history of ${args.request.historyBytes} bytes likely exceeds the context window)`,
|
||||
);
|
||||
}
|
||||
throw new Error(
|
||||
trailerError.code
|
||||
? `Devin stream rejected (${trailerError.code}): ${trailerError.message}`
|
||||
: `Devin stream rejected: ${trailerError.message}`,
|
||||
);
|
||||
}
|
||||
if (!sawEos) throw new Error("Devin stream ended without an EOS trailer");
|
||||
}
|
||||
|
||||
export function streamDevin(
|
||||
model: Model<Api>,
|
||||
context: TranscriptContext,
|
||||
context: Context,
|
||||
options?: SimpleStreamOptions,
|
||||
): AssistantMessageEventStream {
|
||||
const stream = createAssistantMessageEventStream();
|
||||
@@ -398,10 +454,9 @@ export function streamDevin(
|
||||
|
||||
let textOpen = false;
|
||||
let thinkingOpen = false;
|
||||
let toolIndex = -1;
|
||||
let partialJson = "";
|
||||
let toolId = "";
|
||||
let toolName = "";
|
||||
const openTools = new Map<string, number>();
|
||||
const toolJson = new Map<string, string>();
|
||||
let lastToolId: string | undefined;
|
||||
|
||||
const closeText = () => {
|
||||
if (!textOpen) return;
|
||||
@@ -421,38 +476,92 @@ export function streamDevin(
|
||||
}
|
||||
thinkingOpen = false;
|
||||
};
|
||||
const closeTool = () => {
|
||||
if (toolIndex < 0) return;
|
||||
const block = output.content[toolIndex];
|
||||
const closeTool = (id: string) => {
|
||||
const idx = openTools.get(id);
|
||||
if (idx === undefined) return;
|
||||
openTools.delete(id);
|
||||
const block = output.content[idx];
|
||||
if (block.type === "toolCall") {
|
||||
block.arguments = parseStreamingJson(partialJson);
|
||||
const json = toolJson.get(id) ?? "";
|
||||
try {
|
||||
block.arguments = JSON.parse(json);
|
||||
} catch {
|
||||
// keep last parsed object
|
||||
}
|
||||
stream.push({
|
||||
type: "toolcall_end",
|
||||
contentIndex: toolIndex,
|
||||
toolCall: { type: "toolCall", id: toolId, name: toolName, arguments: block.arguments },
|
||||
contentIndex: idx,
|
||||
toolCall: { type: "toolCall", id, name: block.name, arguments: block.arguments },
|
||||
partial: output,
|
||||
});
|
||||
}
|
||||
toolIndex = -1;
|
||||
toolJson.delete(id);
|
||||
if (lastToolId === id) lastToolId = undefined;
|
||||
};
|
||||
const closeAllTools = () => {
|
||||
for (const id of [...openTools.keys()]) closeTool(id);
|
||||
};
|
||||
const currentThinkingBlock = () => {
|
||||
const block = output.content[output.content.length - 1];
|
||||
return block?.type === "thinking" ? block : undefined;
|
||||
};
|
||||
|
||||
try {
|
||||
const apiKey = options?.apiKey;
|
||||
if (!apiKey) throw new Error("No Devin credentials. Run /login devin (uses the local Devin CLI).");
|
||||
if (options?.deferred) throw new Error("The devin provider does not support deferred responses");
|
||||
const host = (options?.env?.DEVIN_API_SERVER_URL || "https://server.codeium.com").replace(/\/$/, "");
|
||||
const modelUid = resolveModelUid(model.id, model.thinkingLevelMap, options?.reasoning);
|
||||
const mapped = mapContextToChat(context);
|
||||
stream.push({ type: "start", partial: output });
|
||||
output.providerThinkingLevel = modelUid;
|
||||
output.responseModel = modelUid;
|
||||
const mapped = mapContextToChat(context, model);
|
||||
const timeoutMs = options?.timeoutMs;
|
||||
const signal =
|
||||
timeoutMs && timeoutMs > 0
|
||||
? anySignal(
|
||||
[options?.signal, AbortSignal.timeout(timeoutMs)].filter((s): s is AbortSignal => Boolean(s)),
|
||||
)
|
||||
: options?.signal;
|
||||
const ids = sessionIds(apiKey, host, options);
|
||||
const userJwt = await getCachedUserJwt(apiKey, host, signal);
|
||||
|
||||
for await (const event of streamChatEvents({
|
||||
const sampling =
|
||||
model.samplingParams || options?.samplingParams
|
||||
? { ...model.samplingParams, ...options?.samplingParams }
|
||||
: {};
|
||||
const wireArgs = {
|
||||
apiKey,
|
||||
host,
|
||||
userJwt,
|
||||
modelUid,
|
||||
systemPrompt: mapped.systemPrompt,
|
||||
messages: mapped.messages,
|
||||
tools: mapped.tools.length > 0 ? mapped.tools : undefined,
|
||||
maxOutputTokens: options?.maxTokens,
|
||||
signal: options?.signal,
|
||||
cascadeId: ids.cascadeId,
|
||||
promptId: randomUUID(),
|
||||
sessionId: ids.sessionId,
|
||||
requestId: BigInt(Date.now()),
|
||||
triggerId: randomUUID(),
|
||||
maxTokens: clampMaxTokensToContext(model, context, options?.maxTokens ?? model.maxTokens),
|
||||
temperature: options?.temperature ?? 1.0,
|
||||
topP: typeof sampling.topP === "number" ? sampling.topP : 0.95,
|
||||
topK: typeof sampling.topK === "number" ? sampling.topK : 40,
|
||||
};
|
||||
const overridden = await options?.onPayload?.(wireArgs, model);
|
||||
const finalArgs =
|
||||
overridden && typeof overridden === "object" ? { ...wireArgs, ...overridden } : wireArgs;
|
||||
const request = buildGetChatMessageRequest(finalArgs);
|
||||
|
||||
stream.push({ type: "start", partial: output });
|
||||
|
||||
for await (const event of streamChatEvents({
|
||||
host,
|
||||
request,
|
||||
headers: options?.headers,
|
||||
signal,
|
||||
fetchFn: options?.fetch,
|
||||
onResponse: options?.onResponse
|
||||
? (response) => options?.onResponse?.(response, model)
|
||||
: undefined,
|
||||
})) {
|
||||
if (event.kind === "text") {
|
||||
closeThinking();
|
||||
@@ -480,46 +589,66 @@ export function streamDevin(
|
||||
block.thinking += event.text;
|
||||
stream.push({ type: "thinking_delta", contentIndex: idx, delta: event.text, partial: output });
|
||||
}
|
||||
} else if (event.kind === "signature") {
|
||||
const block = currentThinkingBlock();
|
||||
if (block) block.thinkingSignature = event.text;
|
||||
} else if (event.kind === "thinking_redacted") {
|
||||
const block = currentThinkingBlock();
|
||||
if (block) block.redacted = true;
|
||||
} else if (event.kind === "message_id") {
|
||||
if (!output.responseId) output.responseId = event.id;
|
||||
} else if (event.kind === "actual_model") {
|
||||
output.responseModel = event.uid;
|
||||
} else if (event.kind === "tool_call_start") {
|
||||
closeText();
|
||||
closeThinking();
|
||||
closeTool();
|
||||
toolId = event.id;
|
||||
toolName = event.name;
|
||||
partialJson = "";
|
||||
output.content.push({ type: "toolCall", id: event.id, name: event.name, arguments: {} });
|
||||
toolIndex = output.content.length - 1;
|
||||
stream.push({ type: "toolcall_start", contentIndex: toolIndex, partial: output });
|
||||
openTools.set(event.id, output.content.length - 1);
|
||||
toolJson.set(event.id, "");
|
||||
lastToolId = event.id;
|
||||
stream.push({ type: "toolcall_start", contentIndex: output.content.length - 1, partial: output });
|
||||
} else if (event.kind === "tool_call_args") {
|
||||
if (toolIndex < 0) continue;
|
||||
partialJson += event.argsDelta;
|
||||
const block = output.content[toolIndex];
|
||||
const id = event.id ?? lastToolId;
|
||||
const idx = id === undefined ? undefined : openTools.get(id);
|
||||
if (id === undefined || idx === undefined) continue;
|
||||
const json = (toolJson.get(id) ?? "") + event.argsDelta;
|
||||
toolJson.set(id, json);
|
||||
const block = output.content[idx];
|
||||
if (block.type === "toolCall") {
|
||||
block.arguments = parseStreamingJson(partialJson);
|
||||
try {
|
||||
block.arguments = JSON.parse(json);
|
||||
} catch {
|
||||
// incomplete json
|
||||
}
|
||||
}
|
||||
stream.push({ type: "toolcall_delta", contentIndex: toolIndex, delta: event.argsDelta, partial: output });
|
||||
stream.push({ type: "toolcall_delta", contentIndex: idx, delta: event.argsDelta, partial: output });
|
||||
} else if (event.kind === "finish") {
|
||||
closeText();
|
||||
closeThinking();
|
||||
closeTool();
|
||||
closeAllTools();
|
||||
if (event.reason === "content_filter") {
|
||||
throw new Error("Devin stop_reason: content_filter");
|
||||
}
|
||||
if (event.reason === "error") {
|
||||
throw new Error("Devin stop_reason: error");
|
||||
}
|
||||
output.stopReason =
|
||||
event.reason === "tool_calls" ? "toolUse" : event.reason === "length" ? "length" : "stop";
|
||||
} else if (event.kind === "usage") {
|
||||
// Empty trailers must not erase usage already received. Metrics are
|
||||
// snapshots, and a frame may contain only some of the components.
|
||||
if (!event.totalTokens) continue;
|
||||
output.usage.input = event.promptTokens ?? output.usage.input;
|
||||
output.usage.output = event.completionTokens ?? output.usage.output;
|
||||
output.usage.cacheRead = event.cachedInputTokens ?? output.usage.cacheRead;
|
||||
output.usage.cacheWrite = event.cacheCreationInputTokens ?? output.usage.cacheWrite;
|
||||
output.usage.totalTokens = output.usage.input + output.usage.output + output.usage.cacheRead + output.usage.cacheWrite;
|
||||
output.usage.input = event.input ?? 0;
|
||||
output.usage.output = event.output ?? 0;
|
||||
output.usage.cacheRead = event.cacheRead ?? 0;
|
||||
output.usage.cacheWrite = event.cacheWrite ?? 0;
|
||||
output.usage.totalTokens =
|
||||
output.usage.input + output.usage.output + output.usage.cacheRead + output.usage.cacheWrite;
|
||||
if (!output.responseModel && event.modelUid) output.responseModel = event.modelUid;
|
||||
calculateCost(model, output.usage);
|
||||
}
|
||||
}
|
||||
|
||||
closeText();
|
||||
closeThinking();
|
||||
closeTool();
|
||||
closeAllTools();
|
||||
stream.push({ type: "done", reason: output.stopReason as "stop" | "length" | "toolUse", message: output });
|
||||
stream.end();
|
||||
} catch (error) {
|
||||
|
||||
Reference in New Issue
Block a user