feat(kb/ask): live SSE streaming of Shira's progress (v0.2.0)

User wanted to see what Shira is doing while she thinks — like Claude
does — instead of staring at an incrementing seconds counter. Now the
ask UI shows one line per agent step as it happens:

  3s  🧠  קוראת את השאלה
  4s  🔍  מחפשת בבסיס הידע: "תקנה 36"
  9s  ✓   נמצאו 5 קטעים רלוונטיים
 10s  🔍  מחפשת בבסיס הידע: "תקנה 36 ביטוח לאומי"
 14s  ✓   נמצאו 5 קטעים רלוונטיים
 15s  🧠  ממשיכה לחקור
 22s  ✍️  מנסחת תשובה
 → תשובה

Client wiring:
- runAsk now opens an EventSource at ?entryPoint=KnowledgeBaseAskStream
  instead of POSTing JSON. Each SSE event of type thinking/tool_start/
  tool_done/tool_error/writing appends a row to a stacked progress log.
  The final {type:answer,text,sources} event triggers renderAskAnswer
  with the existing split view. {type:error} surfaces via showError.
- Module-level state from v0.1.9 unchanged in spirit but now stores the
  EventSource + accumulated events. afterRender on view re-mount replays
  the entire event list and re-binds onmessage/onerror — the SSE
  connection itself doesn't drop because it lives at module scope.
- Elapsed-time pill stays for cadence but is decoration; the real
  signal is the per-event log.

EspoCRM proxy:
- New EntryPoint KnowledgeBaseAskStream — required because EventSource
  is GET-only and ?message= must come via the URL. PHP buffering is
  disabled (zlib + ob + apache_setenv no-gzip), Apache headers are set
  for SSE (no-cache, X-Accel-Buffering: no), and we cURL into
  shira-hermes /kb/ask/stream with CURLOPT_WRITEFUNCTION echoing each
  byte and flushing immediately. The handler ends with exit; to bypass
  EspoCRM's Response wrapper which would otherwise emit headers/body
  on top of our raw stream.
- KnowledgeBaseService gained two public accessors (getStreamingUrl,
  getStreamingApiKey) so the EntryPoint can build the cURL without
  going through request() (which buffers the whole body).
- entryPoints.json registers the new class.

Server (shira-hermes 76fd77f):
- AgentRunner.run accepts an on_event sync callback fired before each
  LLM call (thinking) and around each tool call (tool_start, tool_done,
  tool_error). Hebrew labels in agent_runner._tool_label / _tool_done_label
  ship over the wire ready-to-display — keeps the client free of
  translation concerns.
- POST /kb/ask/stream wraps the existing agent runner in an asyncio
  task feeding an asyncio.Queue, returns StreamingResponse with proper
  SSE headers (X-Accel-Buffering, Cache-Control). 15s comment-line
  heartbeats keep the connection open through proxy idle-timeouts.
- Original POST /kb/ask still works — additive change.

Refs Task Master #8
This commit is contained in:
2026-04-25 13:00:58 +00:00
parent 2fe10727eb
commit a00fc3729c
6 changed files with 307 additions and 63 deletions
+12
View File
@@ -37,6 +37,18 @@
"dependencies": [],
"createdAt": "2026-04-25T10:00:00Z"
},
{
"id": 8,
"title": "feat(kb/ask): SSE streaming with live agent progress",
"description": "User wanted Shira's progress to be visible while she thinks — like Claude does — instead of just an incrementing seconds counter. Implementation: AgentRunner.run accepts an optional sync on_event callback that emits {type, label, ...} dicts at each agent step. New POST /kb/ask/stream returns text/event-stream, with the runner running in an asyncio task whose events feed an asyncio.Queue that the response generator drains as SSE. Client opens EventSource against a new EspoCRM EntryPoint KnowledgeBaseAskStream that proxies the SSE bytes through cURL with output buffering disabled (echo + flush + exit, bypassing the framework Response). The progress UI now stacks one line per event (thinking, tool_start with the query Shira chose, tool_done with the chunk count, writing) instead of just a seconds counter. v0.1.9 _activeAsk persistence pattern preserved — events accumulate at module scope so a view switch + return replays the live log.",
"status": "done",
"priority": "high",
"details": "shira-hermes 76fd77f. EspoCRM EntryPoint registered in metadata/app/entryPoints.json. Hebrew labels live in agent_runner._tool_label / _tool_done_label so the wire format already carries display-ready strings (no client-side mapping). 15s heartbeats keep the connection open through proxy idle-timeouts.",
"testStrategy": "curl -N POST /kb/ask/stream {message:'מהי תקנה 36'} should produce data: lines with thinking → tool_start (with query) → tool_done (with chunk count) → ... → writing → answer. Browser EventSource should drive the same flow into the UI.",
"subtasks": [],
"dependencies": [],
"createdAt": "2026-04-25T13:00:00Z"
},
{
"id": 7,
"title": "fix: search preview duplicates left card for text sources",
@@ -115,7 +115,9 @@ define('modules/knowledge-base/views/kb/index', ['view'], function (Dep) {
const $input = this.$el.find('input[data-name="query"]');
if ($input.length) $input.val(_activeAsk.question);
this.setLoading(true);
this.startAskProgress(_activeAsk.startedAt);
// Pass the whole ask so startAskProgress can replay the
// accumulated event log into the freshly-rendered DOM.
this.startAskProgress(_activeAsk);
this._attachAskHandlers(_activeAsk);
} else if (_lastAsk) {
const $input = this.$el.find('input[data-name="query"]');
@@ -205,93 +207,173 @@ define('modules/knowledge-base/views/kb/index', ['view'], function (Dep) {
_clear(sessionStorage, SS_ASK);
this.setLoading(true);
// Ask goes through the full agent loop (search + rerank + LLM),
// which routinely takes 30-90s. The default jQuery Ajax timeout
// is too short; override it here so the browser doesn't give up
// before shira-hermes responds.
const promise = Espo.Ajax.postRequest('KnowledgeBase/action/ask', {
message: message,
}, {timeout: 240000});
_activeAsk = {question: message, promise: promise, startedAt: Date.now()};
this.startAskProgress(_activeAsk.startedAt);
// Streaming via Server-Sent Events. The browser opens an
// EventSource against KnowledgeBaseAskStream which proxies
// /kb/ask/stream on shira-hermes; one SSE event per agent step
// (thinking, tool_start, tool_done) plus a final answer event.
const url = '?entryPoint=KnowledgeBaseAskStream&message='
+ encodeURIComponent(message);
const es = new EventSource(url);
_activeAsk = {
question: message,
es: es,
events: [],
startedAt: Date.now(),
};
this.startAskProgress(_activeAsk);
this._attachAskHandlers(_activeAsk);
},
// Wire .then/.catch on a (possibly module-level) ask promise to
// whichever view is currently mounted. Called from runAsk for the
// initial bind and from afterRender when the user navigates back to
// the KB view while an earlier request is still in flight.
// Wire SSE handlers on a (possibly module-level) ask. Called from
// runAsk for the initial bind and from afterRender when the user
// navigates back to the KB view while an earlier stream is still
// open. The EventSource itself lives at module scope (in _activeAsk)
// so handlers can be re-bound without dropping the connection.
_attachAskHandlers: function (ask) {
const self = this;
ask.promise.then(res => {
const entry = {
question: ask.question,
text: res.text || '',
sources: res.sources || [],
completedAt: Date.now(),
};
_lastAsk = entry;
_saveJson(sessionStorage, SS_ASK, entry);
if (_activeAsk === ask) _activeAsk = null;
if (self.$el && self.$el.length) {
self.stopAskProgress();
self.setLoading(false);
if (self.mode === 'ask') {
self.renderAskAnswer(entry.text, entry.sources);
ask.es.onmessage = function (msgEvent) {
let data;
try { data = JSON.parse(msgEvent.data); }
catch (e) {
console.warn('KB: malformed SSE payload', e);
return;
}
if (data.type === 'answer') {
self._finishAsk(ask, {
question: ask.question,
text: data.text || '',
sources: data.sources || [],
completedAt: Date.now(),
});
} else if (data.type === 'error') {
self._failAsk(ask, data.message || 'שגיאה בשרת');
} else {
// Status event — record it and append to the live log.
ask.events.push(Object.assign({_clientTs: Date.now()}, data));
if (self.$el && self.$el.length && self.mode === 'ask') {
self._appendAskEvent(ask, ask.events[ask.events.length - 1]);
}
}
}).catch(err => {
if (_activeAsk === ask) _activeAsk = null;
if (self.$el && self.$el.length) {
self.stopAskProgress();
self.setLoading(false);
self.showError(err);
}
});
};
ask.es.onerror = function () {
// Don't tear down on transient blips while the answer is
// already arriving; only treat it as a hard failure if no
// answer has been delivered.
if (ask._completed) return;
self._failAsk(ask, 'הזרם נסגר לפני שהגיעה תשובה');
};
},
startAskProgress: function (startedAt) {
// Show a live elapsed-time indicator so users know the request
// is still running rather than stuck. Typical answers arrive in
// 30-90s; we warn after 2 minutes. `startedAt` is passed in so
// a request that began before a view re-mount keeps showing the
// accumulated elapsed time, not 0.
_finishAsk: function (ask, entry) {
ask._completed = true;
try { ask.es.close(); } catch (e) { /* ignore */ }
_lastAsk = entry;
_saveJson(sessionStorage, SS_ASK, entry);
if (_activeAsk === ask) _activeAsk = null;
if (this.$el && this.$el.length) {
this.stopAskProgress();
this.setLoading(false);
if (this.mode === 'ask') {
this.renderAskAnswer(entry.text, entry.sources);
}
}
},
_failAsk: function (ask, message) {
try { ask.es.close(); } catch (e) { /* ignore */ }
if (_activeAsk === ask) _activeAsk = null;
if (this.$el && this.$el.length) {
this.stopAskProgress();
this.setLoading(false);
this.showError(new Error(message));
}
},
startAskProgress: function (askOrStartedAt) {
// Render a live progress panel: a "thinking" header with elapsed
// time, plus a stacked log of the events Shira's emitted so far
// (one line per LLM step / tool call). `askOrStartedAt` is
// either the active-ask object (from runAsk and afterRender) or
// a bare timestamp (legacy call sites). Re-mounting on view
// switch replays the accumulated event list from ask.events.
if (this.mode !== 'ask') return;
this.stopAskProgress();
startedAt = startedAt || Date.now();
let ask = null;
let startedAt;
if (askOrStartedAt && typeof askOrStartedAt === 'object') {
ask = askOrStartedAt;
startedAt = ask.startedAt || Date.now();
} else {
startedAt = askOrStartedAt || Date.now();
}
const $results = this.$el.find('.kb-results');
$results.html(
'<div class="kb-ask-progress panel panel-info">' +
' <div class="panel-body" style="direction:rtl;text-align:right;">' +
' <i class="fas fa-circle-notch fa-spin" style="margin-left:8px;"></i>' +
' <strong>שירה חושבת…</strong>' +
' <div class="small text-muted" style="margin-top:6px;">' +
' שירה מחפשת בבסיס הידע, מסכמת ומציגה ציטוטים מדויקים.' +
' זה יכול לקחת בין 30 ל-90 שניות.' +
' </div>' +
' <div class="small" style="margin-top:10px;">' +
' <span>זמן שעבר: </span>' +
' <span class="kb-elapsed">0</span>' +
' <span> שניות</span>' +
' <div style="display:flex;align-items:center;gap:8px;margin-bottom:10px;">' +
' <i class="fas fa-circle-notch fa-spin"></i>' +
' <strong>שירה חושבת</strong>' +
' <span class="text-muted small kb-elapsed-pill" ' +
'style="margin-right:auto;font-variant-numeric:tabular-nums;">0 שניות</span>' +
' </div>' +
' <div class="kb-events" style="' +
'border-top:1px solid #d8e0e8;padding-top:8px;' +
'max-height:60vh;overflow-y:auto;font-size:0.95em;"></div>' +
' </div>' +
'</div>'
);
// Replay any events already collected (matters when the user
// navigates away mid-think and comes back to a still-streaming
// ask — the stream kept going at module scope, but the DOM was
// freshly re-rendered).
if (ask && Array.isArray(ask.events)) {
ask.events.forEach(ev => this._appendAskEvent(ask, ev));
}
const tick = () => {
const sec = Math.round((Date.now() - startedAt) / 1000);
const $elapsed = this.$el.find('.kb-elapsed');
if ($elapsed.length) $elapsed.text(sec);
if (sec === 120) {
this.$el.find('.kb-ask-progress .panel-body').append(
'<div class="alert alert-warning" style="margin-top:10px;">' +
'לוקח יותר זמן מהרגיל. עדיין ממתינים…</div>'
);
}
this.$el.find('.kb-elapsed-pill').text(sec + ' שניות');
};
tick();
this._askTimer = setInterval(tick, 1000);
},
_appendAskEvent: function (ask, ev) {
const $events = this.$el.find('.kb-events');
if (!$events.length) return;
const startedAt = (ask && ask.startedAt) || Date.now();
const ts = ev._clientTs || Date.now();
const sec = Math.max(0, Math.round((ts - startedAt) / 1000));
const icon = {
thinking: '🧠',
tool_start: '🔍',
tool_done: '✓',
tool_error: '⚠',
writing: '✍️',
}[ev.type] || '·';
const colour = {
thinking: '#475569',
tool_start: '#0f5298',
tool_done: '#1e6f40',
tool_error: '#a02622',
writing: '#5a3a91',
}[ev.type] || '#475569';
const label = this.escape(ev.label || ev.type);
const row = $(
'<div style="padding:3px 0;color:' + colour + ';display:flex;gap:8px;align-items:flex-start;">' +
' <span style="display:inline-block;min-width:40px;color:#94a3b8;font-variant-numeric:tabular-nums;">' +
sec + 'ש</span>' +
' <span style="margin:0 2px;">' + icon + '</span>' +
' <span style="flex:1 1 auto;word-break:break-word;">' + label + '</span>' +
'</div>'
);
$events.append(row);
// Always keep the latest line visible.
$events.scrollTop($events[0].scrollHeight);
},
stopAskProgress: function () {
if (this._askTimer) {
clearInterval(this._askTimer);
@@ -0,0 +1,128 @@
<?php
namespace Espo\Modules\KnowledgeBase\EntryPoints;
use Espo\Core\Api\Request;
use Espo\Core\Api\Response;
use Espo\Core\EntryPoint\EntryPoint;
use Espo\Core\Exceptions\BadRequest;
use Espo\Core\Exceptions\Error;
use Espo\Core\Exceptions\Forbidden;
use Espo\Core\InjectableFactory;
use Espo\Entities\User;
use Espo\Modules\KnowledgeBase\Services\KnowledgeBaseService;
/**
* Streams Server-Sent Events from shira-hermes /kb/ask/stream to the
* browser's EventSource. The browser opens this with a GET (EventSource
* has no POST), passes the question via ?message=, and we forward it
* to shira-hermes with the firm-wide API key — the credential never
* leaves the EspoCRM server.
*
* Response is text/event-stream with no buffering. We bypass EspoCRM's
* Response wrapper entirely (echo + flush + exit) because the framework
* is wired for buffered JSON responses, not infinite streams.
*
* Access: `?entryPoint=KnowledgeBaseAskStream&message=<urlencoded>`
*/
class KnowledgeBaseAskStream implements EntryPoint
{
public function __construct(
private InjectableFactory $injectableFactory,
private User $user,
) {}
public function run(Request $request, Response $response): void
{
if ($this->user->isPortal()) {
throw new Forbidden('Portal users have no KB access.');
}
$message = $request->getQueryParam('message');
if (!is_string($message) || trim($message) === '') {
throw new BadRequest('message (string) is required.');
}
if (strlen($message) > 2000) {
throw new BadRequest('message is too long.');
}
$service = $this->injectableFactory->create(KnowledgeBaseService::class);
try {
$upstreamUrl = $service->getStreamingUrl('/kb/ask/stream');
$apiKey = $service->getStreamingApiKey();
} catch (\Throwable $e) {
throw new Error('SmartAssistant integration is not configured: ' . $e->getMessage());
}
// Disable PHP/Apache output buffering and compression so each SSE
// event flushes to the wire as soon as it arrives from upstream.
// Order matters: clean any open buffers BEFORE setting headers.
while (ob_get_level() > 0) {
@ob_end_clean();
}
@ini_set('zlib.output_compression', 'Off');
@ini_set('output_buffering', 'Off');
@ini_set('implicit_flush', '1');
if (function_exists('apache_setenv')) {
@apache_setenv('no-gzip', '1');
}
ignore_user_abort(false);
@set_time_limit(300);
header('Content-Type: text/event-stream; charset=utf-8');
header('Cache-Control: no-cache, no-transform');
header('X-Accel-Buffering: no');
header('Connection: keep-alive');
// Defense in depth: same locked-down CSP we use for the PDF entry.
header("Content-Security-Policy: default-src 'self'");
$payload = json_encode(['message' => $message], JSON_UNESCAPED_UNICODE);
$ch = curl_init();
curl_setopt_array($ch, [
CURLOPT_URL => $upstreamUrl,
CURLOPT_POST => true,
CURLOPT_POSTFIELDS => $payload,
CURLOPT_HTTPHEADER => [
'Content-Type: application/json',
'Accept: text/event-stream',
'X-Api-Key: ' . $apiKey,
],
CURLOPT_RETURNTRANSFER => false,
CURLOPT_BUFFERSIZE => 1024,
CURLOPT_TIMEOUT => 300,
CURLOPT_CONNECTTIMEOUT => 10,
CURLOPT_WRITEFUNCTION => function ($ch, $data) {
if (connection_aborted()) {
// Browser closed the EventSource — bail so cURL stops.
return -1;
}
echo $data;
@ob_flush();
@flush();
return strlen($data);
},
]);
$ok = curl_exec($ch);
$errno = curl_errno($ch);
$errmsg = curl_error($ch);
curl_close($ch);
if (!$ok && $errno !== 23 /* aborted by callback */) {
// Best effort: if upstream errored before any data was
// flushed, surface a final SSE error event so the client
// doesn't silently spin.
echo "data: " . json_encode(
['type' => 'error', 'message' => 'upstream: ' . $errmsg],
JSON_UNESCAPED_UNICODE,
) . "\n\n";
@ob_flush();
@flush();
}
// We've fully written the response. Bypass EspoCRM's Response
// emit() to avoid double-headers / extra buffering.
exit;
}
}
@@ -1,5 +1,8 @@
{
"KnowledgeBasePdf": {
"className": "Espo\\Modules\\KnowledgeBase\\EntryPoints\\KnowledgeBasePdf"
},
"KnowledgeBaseAskStream": {
"className": "Espo\\Modules\\KnowledgeBase\\EntryPoints\\KnowledgeBaseAskStream"
}
}
@@ -140,6 +140,25 @@ class KnowledgeBaseService
return $data->apiKey ?? null;
}
/**
* Public accessors so the streaming EntryPoint can build a cURL
* connection to shira-hermes without going through the JSON request()
* helper (which buffers the whole body).
*/
public function getStreamingUrl(string $path): string
{
return rtrim($this->getBaseUrl(), '/') . $path;
}
public function getStreamingApiKey(): string
{
$key = $this->getApiKey();
if (!$key) {
throw new Error('SmartAssistant API key is not configured.');
}
return $key;
}
private function get(string $path): array
{
return $this->request('GET', $path, null);
+1 -1
View File
@@ -1,7 +1,7 @@
{
"name": "KnowledgeBase",
"module": "KnowledgeBase",
"version": "0.1.11",
"version": "0.2.0",
"acceptableVersions": [
">=8.0.0"
],