let5see/clawdbot
1
0
Fork 0
Code Issues Pull Requests Actions 4 Packages Projects Releases Wiki Activity

feat: make block streaming break configurable

Browse Source
This commit is contained in:
Peter Steinberger
2026-01-03 00:52:02 +01:00
parent ea7d967625
commit 2d28fa34f5
8 changed files with 138 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
CHANGELOG.md
View File

@@ -54,7 +54,7 @@
### Fixes
- Chat UI: keep the chat scrolled to the latest message after switching sessions.
- Auto-reply: stream completed reply blocks as soon as they finish (configurable default); skip empty tool-only blocks unless verbose.
- Auto-reply: stream completed reply blocks as soon as they finish (configurable default + break); skip empty tool-only blocks unless verbose.
- CLI onboarding: persist gateway token in config so local CLI auth works; recommend auth Off unless you need multi-machine access.
- Control UI: accept a `?token=` URL param to auto-fill Gateway auth; onboarding now opens the dashboard with token auth when configured.
- Agent prompt: remove hardcoded user name in system prompt example.

1
docs/agent.md
View File

@@ -73,6 +73,7 @@ Legacy Pi/Tau session folders are **not** read.
Incoming user messages are queued while the agent is streaming. The queue is checked **after each tool call**. If a queued message is present, remaining tool calls from the current assistant message are skipped (error tool results with "Skipped due to queued user message."), then the queued user message is injected before the next assistant response.
Block streaming sends completed assistant blocks as soon as they finish; disable
via `agent.blockStreamingDefault: "off"` if you only want the final response.
Tune the boundary via `agent.blockStreamingBreak` (`text_end` vs `message_end`).
## Configuration (minimal)

11
docs/configuration.md
View File

@@ -331,6 +331,7 @@ Controls the embedded agent runtime (model/thinking/verbose/timeouts).
thinkingDefault: "low",
verboseDefault: "off",
blockStreamingDefault: "on",
blockStreamingBreak: "text_end",
timeoutSeconds: 600,
mediaMaxMb: 5,
heartbeat: {
@@ -355,9 +356,13 @@ deprecation fallback.
Z.AI models are available as `zai/<model>` (e.g. `zai/glm-4.7`) and require
`ZAI_API_KEY` (or legacy `Z_AI_API_KEY`) in the environment.
`agent.blockStreamingDefault` controls whether completed assistant blocks
(`message_end` chunks) are sent immediately (default: `on`). Set to `off` to
only deliver the final consolidated reply.
`agent.blockStreamingDefault` controls whether completed assistant blocks are
sent immediately (default: `on`). Set to `off` to only deliver the final
consolidated reply.
`agent.blockStreamingBreak` controls what “block” means:
- `text_end` (default): end of each assistant text content block (before tool calls)
- `message_end`: end of the whole assistant message (may wait across tools)
`agent.heartbeat` configures periodic heartbeat runs:
- `every`: duration string (`ms`, `s`, `m`, `h`); default unit minutes. Omit or set

2
src/agents/pi-embedded-runner.ts
View File

@@ -331,6 +331,7 @@ export async function runEmbeddedPiAgent(params: {
text?: string;
mediaUrls?: string[];
}) => void | Promise<void>;
blockReplyBreak?: "text_end" | "message_end";
onToolResult?: (payload: {
text?: string;
mediaUrls?: string[];
@@ -494,6 +495,7 @@ export async function runEmbeddedPiAgent(params: {
shouldEmitToolResult: params.shouldEmitToolResult,
onToolResult: params.onToolResult,
onBlockReply: params.onBlockReply,
blockReplyBreak: params.blockReplyBreak,
onPartialReply: params.onPartialReply,
onAgentEvent: params.onAgentEvent,
enforceFinalTag: params.enforceFinalTag,

54
src/agents/pi-embedded-subscribe.test.ts
View File

@@ -114,6 +114,7 @@ describe("subscribeEmbeddedPiSession", () => {
>[0]["session"],
runId: "run",
onBlockReply,
blockReplyBreak: "message_end",
});
const assistantMessage = {
@@ -128,6 +129,59 @@ describe("subscribeEmbeddedPiSession", () => {
expect(payload.text).toBe("Hello block");
});
it("emits block replies on text_end and does not duplicate on message_end", () => {
let handler: ((evt: unknown) => void) | undefined;
const session: StubSession = {
subscribe: (fn) => {
handler = fn;
return () => {};
},
};
const onBlockReply = vi.fn();
const subscription = subscribeEmbeddedPiSession({
session: session as unknown as Parameters<
typeof subscribeEmbeddedPiSession
>[0]["session"],
runId: "run",
onBlockReply,
blockReplyBreak: "text_end",
});
handler?.({
type: "message_update",
message: { role: "assistant" },
assistantMessageEvent: {
type: "text_delta",
delta: "Hello block",
},
});
handler?.({
type: "message_update",
message: { role: "assistant" },
assistantMessageEvent: {
type: "text_end",
},
});
expect(onBlockReply).toHaveBeenCalledTimes(1);
const payload = onBlockReply.mock.calls[0][0];
expect(payload.text).toBe("Hello block");
expect(subscription.assistantTexts).toEqual(["Hello block"]);
const assistantMessage = {
role: "assistant",
content: [{ type: "text", text: "Hello block" }],
} as AssistantMessage;
handler?.({ type: "message_end", message: assistantMessage });
expect(onBlockReply).toHaveBeenCalledTimes(1);
expect(subscription.assistantTexts).toEqual(["Hello block"]);
});
it("waits for auto-compaction retry and clears buffered text", async () => {
const listeners: SessionEventHandler[] = [];
const session = {

85
src/agents/pi-embedded-subscribe.ts
View File

@@ -61,6 +61,7 @@ export function subscribeEmbeddedPiSession(params: {
text?: string;
mediaUrls?: string[];
}) => void | Promise<void>;
blockReplyBreak?: "text_end" | "message_end";
onPartialReply?: (payload: {
text?: string;
mediaUrls?: string[];
@@ -74,8 +75,10 @@ export function subscribeEmbeddedPiSession(params: {
const assistantTexts: string[] = [];
const toolMetas: Array<{ toolName?: string; meta?: string }> = [];
const toolMetaById = new Map<string, string | undefined>();
const blockReplyBreak = params.blockReplyBreak ?? "text_end";
let deltaBuffer = "";
let lastStreamedAssistant: string | undefined;
let assistantTextBaseline = 0;
let compactionInFlight = false;
let pendingCompactionRetry = 0;
let compactionRetryResolve: (() => void) | undefined;
@@ -149,6 +152,7 @@ export function subscribeEmbeddedPiSession(params: {
toolMetaById.clear();
deltaBuffer = "";
lastStreamedAssistant = undefined;
assistantTextBaseline = 0;
toolDebouncer.flush();
};
@@ -264,38 +268,55 @@ export function subscribeEmbeddedPiSession(params: {
: "";
if (chunk) {
deltaBuffer += chunk;
const cleaned = params.enforceFinalTag
? stripThinkingSegments(stripUnpairedThinkingTags(deltaBuffer))
: stripThinkingSegments(deltaBuffer);
const next = params.enforceFinalTag
? (extractFinalText(cleaned)?.trim() ?? cleaned.trim())
: cleaned.trim();
if (next && next !== lastStreamedAssistant) {
lastStreamedAssistant = next;
}
const cleaned = params.enforceFinalTag
? stripThinkingSegments(stripUnpairedThinkingTags(deltaBuffer))
: stripThinkingSegments(deltaBuffer);
const next = params.enforceFinalTag
? (extractFinalText(cleaned)?.trim() ?? cleaned.trim())
: cleaned.trim();
if (next && next !== lastStreamedAssistant) {
lastStreamedAssistant = next;
const { text: cleanedText, mediaUrls } =
splitMediaFromOutput(next);
emitAgentEvent({
runId: params.runId,
stream: "assistant",
data: {
text: cleanedText,
mediaUrls: mediaUrls?.length ? mediaUrls : undefined,
},
});
params.onAgentEvent?.({
stream: "assistant",
data: {
text: cleanedText,
mediaUrls: mediaUrls?.length ? mediaUrls : undefined,
},
});
if (params.onPartialReply) {
void params.onPartialReply({
text: cleanedText,
mediaUrls: mediaUrls?.length ? mediaUrls : undefined,
});
}
}
if (evtType === "text_end" && blockReplyBreak === "text_end") {
if (next) assistantTexts.push(next);
if (next && params.onBlockReply) {
const { text: cleanedText, mediaUrls } =
splitMediaFromOutput(next);
emitAgentEvent({
runId: params.runId,
stream: "assistant",
data: {
text: cleanedText,
mediaUrls: mediaUrls?.length ? mediaUrls : undefined,
},
});
params.onAgentEvent?.({
stream: "assistant",
data: {
text: cleanedText,
mediaUrls: mediaUrls?.length ? mediaUrls : undefined,
},
});
if (params.onPartialReply) {
void params.onPartialReply({
if (cleanedText || (mediaUrls && mediaUrls.length > 0)) {
void params.onBlockReply({
text: cleanedText,
mediaUrls: mediaUrls?.length ? mediaUrls : undefined,
});
}
}
deltaBuffer = "";
lastStreamedAssistant = undefined;
}
}
}
@@ -317,8 +338,17 @@ export function subscribeEmbeddedPiSession(params: {
params.enforceFinalTag && cleaned
? (extractFinalText(cleaned)?.trim() ?? cleaned)
: cleaned;
if (text) assistantTexts.push(text);
if (text && params.onBlockReply) {
const addedDuringMessage =
assistantTexts.length > assistantTextBaseline;
if (!addedDuringMessage && text) assistantTexts.push(text);
assistantTextBaseline = assistantTexts.length;
if (
blockReplyBreak === "message_end" &&
text &&
params.onBlockReply
) {
const { text: cleanedText, mediaUrls } = splitMediaFromOutput(text);
if (cleanedText || (mediaUrls && mediaUrls.length > 0)) {
void params.onBlockReply({
@@ -328,6 +358,7 @@ export function subscribeEmbeddedPiSession(params: {
}
}
deltaBuffer = "";
lastStreamedAssistant = undefined;
}
}

5
src/auto-reply/reply.ts
View File

@@ -604,6 +604,10 @@ export async function getReplyFromConfig(
(agentCfg?.verboseDefault as VerboseLevel | undefined);
const resolvedBlockStreaming =
agentCfg?.blockStreamingDefault === "off" ? "off" : "on";
const resolvedBlockStreamingBreak =
agentCfg?.blockStreamingBreak === "message_end"
? "message_end"
: "text_end";
const blockStreamingEnabled = resolvedBlockStreaming === "on";
const streamedPayloadKeys = new Set<string>();
const pendingBlockTasks = new Set<Promise<void>>();
@@ -1368,6 +1372,7 @@ export async function getReplyFromConfig(
verboseLevel: resolvedVerboseLevel,
timeoutMs,
runId,
blockReplyBreak: resolvedBlockStreamingBreak,
onPartialReply: opts?.onPartialReply
? async (payload) => {
let text = payload.text;

9
src/config/config.ts
View File

@@ -526,6 +526,12 @@ export type ClawdisConfig = {
verboseDefault?: "off" | "on";
/** Default block streaming level when no override is present. */
blockStreamingDefault?: "off" | "on";
/**
* Block streaming boundary:
* - "text_end": end of each assistant text content block (before tool calls)
* - "message_end": end of the whole assistant message (may include tool blocks)
*/
blockStreamingBreak?: "text_end" | "message_end";
timeoutSeconds?: number;
/** Max inbound media size in MB for agent-visible attachments (text note or future image attach). */
mediaMaxMb?: number;
@@ -906,6 +912,9 @@ const ClawdisSchema = z.object({
blockStreamingDefault: z
.union([z.literal("off"), z.literal("on")])
.optional(),
blockStreamingBreak: z
.union([z.literal("text_end"), z.literal("message_end")])
.optional(),
timeoutSeconds: z.number().int().positive().optional(),
mediaMaxMb: z.number().positive().optional(),
typingIntervalSeconds: z.number().int().positive().optional(),