ui/ux: hide noisy memory meter, surface accurate post-compact tokens

Drop the inline 'Memory N% full' pill from AgentChat and the mode label from AgentCard so the chat header stays quiet until tools are connected. Backend now emits agent:context_update after compact and clear so the session token counters reflect the trimmed history instead of stale pre-compact totals, with a new _estimate_post_compact_input helper and matching invariants in test_v2_invariants.
2026-09-12 04:37:44 +02:00 · 2026-06-13 00:18:11 -07:00
parent d1e3c37b27
commit 392a813bb3
8 changed files with 172 additions and 37 deletions
@@ -53,6 +53,7 @@ from backend.apps.agents.manager.prompt.tool_catalog import (
 from backend.apps.agents.core.aux_llm import _safe_resp_text, clean_short_label
 from backend.apps.agents.manager.session.history_compaction import (
    _build_history_prefix,
+    _estimate_post_compact_input,
    _get_branch_messages,
    _truncate_large_tool_result,
 )
@@ -377,6 +378,35 @@ class AgentManager:
        session.compacted_through_msg_id = last_id
        return True

+    async def _emit_context_update(
+        self,
+        session_id: str,
+        session: AgentSession,
+        *,
+        input_tokens: int | None = None,
+        output_tokens: int | None = None,
+        cache_read_tokens: int = 0,
+        cache_read_pct: float = 0.0,
+    ) -> None:
+        if input_tokens is None:
+            input_tokens = int(session.tokens.get("input", 0) or 0)
+        if output_tokens is None:
+            output_tokens = int(session.tokens.get("output", 0) or 0)
+        session.tokens["input"] = input_tokens
+        session.tokens["output"] = output_tokens
+        ctx_window = max(1, getattr(session, "context_window", 0) or 200_000)
+        await ws_manager.send_to_session(session_id, "agent:context_update", {
+            "session_id": session_id,
+            "input_tokens": input_tokens,
+            "output_tokens": output_tokens,
+            "cache_read_tokens": cache_read_tokens,
+            "cache_read_pct": cache_read_pct,
+            "ctx_used_pct": round(input_tokens / ctx_window, 4) if input_tokens else 0.0,
+            "context_window": ctx_window,
+            "framework_overhead_tokens": session.framework_overhead_tokens,
+            "active_mcps": list(session.active_mcps),
+        })
+
    def _build_prompt_content(self, prompt: str, images: list | None = None, context_paths: list | None = None, forced_tools: list[str] | None = None, attached_skills: list | None = None, api_type: str = "anthropic", model: str = ""):
        return _build_prompt_content(prompt, images, context_paths, forced_tools, attached_skills, api_type, model)

@@ -1813,11 +1843,18 @@ class AgentManager:
            # zero latency on the user's turn.
            try:
                if self._maybe_compact(session):
+                    new_input = _estimate_post_compact_input(session)
                    await ws_manager.send_to_session(session_id, "agent:context_status", {
                        "session_id": session_id,
                        "reason": "compacted",
                        "compacted_through_msg_id": session.compacted_through_msg_id,
                    })
+                    await self._emit_context_update(
+                        session_id,
+                        session,
+                        input_tokens=new_input,
+                        output_tokens=session.tokens.get("output", 0),
+                    )
            except Exception:
                logger.exception("compaction failed; proceeding without it")

@@ -2,6 +2,7 @@ from backend.config.Apps import SubApp
 from backend.apps.agents.agent_manager import agent_manager
 from backend.apps.agents.core.ws_manager import ws_manager
 from backend.apps.agents.core.models import AgentConfig, ApprovalResponse
+from backend.apps.agents.manager.session.history_compaction import _estimate_post_compact_input
 from contextlib import asynccontextmanager
 from fastapi import WebSocket, WebSocketDisconnect, HTTPException
 from fastapi.responses import JSONResponse
@@ -304,6 +305,12 @@ async def compact_session(session_id: str):
                "reason": "compacted",
                "compacted_through_msg_id": session.compacted_through_msg_id,
            })
+            await agent_manager._emit_context_update(
+                session_id,
+                session,
+                input_tokens=_estimate_post_compact_input(session),
+                output_tokens=session.tokens.get("output", 0),
+            )
        except Exception:
            pass
    return {"ok": True, "compacted": fired}
@@ -329,6 +336,12 @@ async def clear_session(session_id: str):
            "status": session.status,
            "session": session.model_dump(mode="json"),
        })
+        await agent_manager._emit_context_update(
+            session_id,
+            session,
+            input_tokens=0,
+            output_tokens=0,
+        )
    except Exception:
        pass
    return {"ok": True}
@@ -817,4 +830,3 @@ async def subscriptions_disconnect(body: dict):
        return {"ok": False, "error": "Connection not found"}
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))
-
@@ -71,6 +71,39 @@ def _build_history_prefix(messages, cutoff_msg_id: str | None = None) -> str:
    return "<prior_conversation>\n" + "\n".join(lines) + "\n</prior_conversation>"


+def _estimate_post_compact_input(session) -> int:
+    """Return a conservative token estimate after compaction trims history."""
+    try:
+        messages = _get_branch_messages(session)
+        cutoff_msg_id = getattr(session, "compacted_through_msg_id", None)
+        if cutoff_msg_id:
+            skip_idx = next(
+                (i for i, m in enumerate(messages) if m.id == cutoff_msg_id),
+                -1,
+            )
+            if skip_idx >= 0:
+                messages = messages[skip_idx + 1:]
+        surviving_chars = 0
+        for message in messages:
+            if getattr(message, "hidden", False):
+                continue
+            content = getattr(message, "content", "")
+            if isinstance(content, str):
+                serialized = content
+            else:
+                try:
+                    serialized = json.dumps(content, ensure_ascii=False)
+                except Exception:
+                    serialized = str(content)
+            surviving_chars += len(serialized)
+        framework_overhead = int(getattr(session, "framework_overhead_tokens", 0) or 0)
+        summary_overhead = 200 if cutoff_msg_id else 0
+        return max(0, framework_overhead + summary_overhead + (surviving_chars // 4))
+    except Exception:
+        logger.debug("post-compact token estimate failed", exc_info=True)
+        return max(0, int(getattr(session, "framework_overhead_tokens", 0) or 0))
+
+
 def _truncate_large_tool_result(content: object, session_id: str, msg_id: str, max_bytes: int = 50_000) -> tuple[object, str | None]:
    """Spill a large tool_result body to disk, return a truncated
    inline replacement plus the on-disk path (or None if untouched).
@@ -784,6 +784,66 @@ def test_compact_threshold_default():
    assert s.context_window == 200_000


+def test_post_compact_estimate_excludes_compacted_messages():
+    from backend.apps.agents.core.models import AgentSession, Message
+    from backend.apps.agents.manager.session.history_compaction import (
+        _estimate_post_compact_input,
+    )
+
+    messages = [
+        Message(id=f"m{i}", role="user", content=("old" * 1000 if i < 6 else "keep"))
+        for i in range(8)
+    ]
+    s = AgentSession(id="x", name="t", model="sonnet", mode="agent")
+    s.messages = messages
+    s.compacted_through_msg_id = "m5"
+    s.framework_overhead_tokens = 100
+
+    assert _estimate_post_compact_input(s) == 100 + 200 + (len("keepkeep") // 4)
+
+
+@pytest.mark.asyncio
+async def test_context_update_emitter_refreshes_session_tokens(monkeypatch):
+    import backend.apps.agents.agent_manager as agent_manager_module
+    from backend.apps.agents.agent_manager import AgentManager
+    from backend.apps.agents.core.models import AgentSession
+
+    sent = []
+
+    async def fake_send_to_session(session_id, event, payload):
+        sent.append((session_id, event, payload))
+
+    monkeypatch.setattr(
+        agent_manager_module.ws_manager,
+        "send_to_session",
+        fake_send_to_session,
+    )
+    s = AgentSession(id="x", name="t", model="sonnet", mode="agent")
+    s.context_window = 1_000
+    s.tokens = {"input": 900, "output": 7}
+    s.framework_overhead_tokens = 42
+    s.active_mcps = ["github"]
+
+    await AgentManager()._emit_context_update("x", s, input_tokens=250)
+
+    assert s.tokens == {"input": 250, "output": 7}
+    assert sent == [(
+        "x",
+        "agent:context_update",
+        {
+            "session_id": "x",
+            "input_tokens": 250,
+            "output_tokens": 7,
+            "cache_read_tokens": 0,
+            "cache_read_pct": 0.0,
+            "ctx_used_pct": 0.25,
+            "context_window": 1_000,
+            "framework_overhead_tokens": 42,
+            "active_mcps": ["github"],
+        },
+    )]
+
+
 # ===========================================================================
 # Group L, Sentence-case display (the parseMcpToolName fix)
 # ===========================================================================
@@ -1,12 +1,12 @@
 {
  "name": "openswarm",
-  "version": "1.1.70",
+  "version": "1.2.77",
  "lockfileVersion": 3,
  "requires": true,
  "packages": {
    "": {
      "name": "openswarm",
-      "version": "1.1.70",
+      "version": "1.2.77",
      "hasInstallScript": true,
      "dependencies": {
        "electron-updater": "6.8.3",
@@ -567,6 +567,7 @@
      "integrity": "sha512-fgFx7Hfoq60ytK2c7DhnF8jIvzYgOMxfugjLOSMHjLIPgenqa7S7oaagATUq99mV6IYvN2tRmC0wnTYX6iPbMw==",
      "dev": true,
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "fast-deep-equal": "^3.1.1",
        "fast-json-stable-stringify": "^2.0.0",
@@ -1434,6 +1435,7 @@
      "integrity": "sha512-glMJgnTreo8CFINujtAhCgN96QAqApDMZ8Vl1r8f0QT8QprvC1UCltV4CcWj20YoIyLZx6IUskaJZ0NV8fokcg==",
      "dev": true,
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "app-builder-lib": "26.8.1",
        "builder-util": "26.8.1",
@@ -1582,6 +1584,7 @@
      "integrity": "sha512-o288fIdgPLHA76eDrFADHPoo7VyGkDCYbLV1GzndaMSAVBoZrGvM9m2IehdcVMzdAZJ2eV9bgyissQXHv5tGzA==",
      "dev": true,
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "app-builder-lib": "26.8.1",
        "builder-util": "26.8.1",
@@ -2885,6 +2888,7 @@
      "integrity": "sha512-QP88BAKvMam/3NxH6vj2o21R6MjxZUAd6nlwAS/pnGvN9IVLocLHxGYIzFhg6fUQ+5th6P4dv4eW9jX3DSIj7A==",
      "dev": true,
      "license": "MIT",
+      "peer": true,
      "engines": {
        "node": ">=12"
      },
@@ -84,6 +84,7 @@
      "integrity": "sha512-CGOfOJqWjg2qW/Mb6zNsDm+u5vFQ8DxXfbM09z69p5Z6+mE1ikP2jUXw+j42Pf1XTYED2Rni5f95npYeuwMDQA==",
      "dev": true,
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "@babel/code-frame": "^7.29.0",
        "@babel/generator": "^7.29.0",
@@ -1964,6 +1965,7 @@
      "resolved": "https://registry.npmjs.org/@emotion/react/-/react-11.14.0.tgz",
      "integrity": "sha512-O000MLDBDdk/EohJPFUqvnp4qnHeYkVP5B0xEG0D/L7cOKP9kefu2DXn8dj74cQfsEzUqh+sr1RzFqiL1o+PpA==",
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "@babel/runtime": "^7.18.3",
        "@emotion/babel-plugin": "^11.13.5",
@@ -2007,6 +2009,7 @@
      "resolved": "https://registry.npmjs.org/@emotion/styled/-/styled-11.14.1.tgz",
      "integrity": "sha512-qEEJt42DuToa3gurlH4Qqc1kVpNq8wO8cJtDzU46TjlzWjDlsVyevtYCRijVq3SrHsROS+gVQ8Fnea108GnKzw==",
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "@babel/runtime": "^7.18.3",
        "@emotion/babel-plugin": "^11.13.5",
@@ -2242,6 +2245,7 @@
      "resolved": "https://registry.npmjs.org/@mui/material/-/material-7.3.10.tgz",
      "integrity": "sha512-cHvGOk2ZEfbQt3LnGe0ZKd/ETs9gsUpkW66DCO+GSjMZhpdKU4XsuIr7zJ/B/2XaN8ihxuzHfYAR4zPtCN4RYg==",
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "@babel/runtime": "^7.28.6",
        "@mui/core-downloads-tracker": "^7.3.10",
@@ -3374,6 +3378,7 @@
      "resolved": "https://registry.npmjs.org/@types/react/-/react-18.3.28.tgz",
      "integrity": "sha512-z9VXpC7MWrhfWipitjNdgCauoMLRdIILQsAEV+ZesIzBq/oUlxk0m3ApZuMFCXdnS4U7KrI+l3WRUEGQ8K1QKw==",
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "@types/prop-types": "*",
        "csstype": "^3.2.2"
@@ -3770,6 +3775,7 @@
      "integrity": "sha512-UVJyE9MttOsBQIDKw1skb9nAwQuR5wuGD3+82K6JgJlm/Y+KI92oNsMNGZCYdDsVtRHSak0pcV5Dno5+4jh9sw==",
      "dev": true,
      "license": "MIT",
+      "peer": true,
      "bin": {
        "acorn": "bin/acorn"
      },
@@ -3809,6 +3815,7 @@
      "integrity": "sha512-PlXPeEWMXMZ7sPYOHqmDyCJzcfNrUr3fGNKtezX14ykXOEIvyK81d+qydx89KY5O71FKMPaQ2vBfBFI5NHR63A==",
      "dev": true,
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "fast-deep-equal": "^3.1.3",
        "fast-uri": "^3.0.1",
@@ -4137,6 +4144,7 @@
        }
      ],
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "baseline-browser-mapping": "^2.10.12",
        "caniuse-lite": "^1.0.30001782",
@@ -8184,6 +8192,7 @@
      "integrity": "sha512-QP88BAKvMam/3NxH6vj2o21R6MjxZUAd6nlwAS/pnGvN9IVLocLHxGYIzFhg6fUQ+5th6P4dv4eW9jX3DSIj7A==",
      "dev": true,
      "license": "MIT",
+      "peer": true,
      "engines": {
        "node": ">=12"
      },
@@ -8240,6 +8249,7 @@
        }
      ],
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "nanoid": "^3.3.11",
        "picocolors": "^1.1.1",
@@ -8458,6 +8468,7 @@
      "resolved": "https://registry.npmjs.org/react/-/react-18.3.1.tgz",
      "integrity": "sha512-wS+hAgJShR0KhEvPJArfuPVN1+Hz1t0Y6n5jLrGQbkb4urgPE/0Rve+1kMB1v/oWgHgm4WIcV+i7F2pTVj+2iQ==",
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "loose-envify": "^1.1.0"
      },
@@ -8470,6 +8481,7 @@
      "resolved": "https://registry.npmjs.org/react-dom/-/react-dom-18.3.1.tgz",
      "integrity": "sha512-5m4nQKp+rZRb09LNH59GM4BxTh9251/ylbKIbpe7TpGxfJ+9kv6BLkLBXIjjspbgbnIBNqlI23tRnTWT0snUIw==",
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "loose-envify": "^1.1.0",
        "scheduler": "^0.23.2"
@@ -8516,6 +8528,7 @@
      "resolved": "https://registry.npmjs.org/react-redux/-/react-redux-9.2.0.tgz",
      "integrity": "sha512-ROY9fvHhwOD9ySfrF0wmvu//bKCQ6AeZZq1nJNtbDC+kk5DuSuNX/n6YWYF/SYy7bSba4D4FSz8DJeKY/S/r+g==",
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "@types/use-sync-external-store": "^0.0.6",
        "use-sync-external-store": "^1.4.0"
@@ -8654,7 +8667,8 @@
      "version": "5.0.1",
      "resolved": "https://registry.npmjs.org/redux/-/redux-5.0.1.tgz",
      "integrity": "sha512-M9/ELqF6fy8FwmkpnF0S3YKOqMyoWJ4+CS5Efg2ct3oY9daQvd/Pc71FpGZsVsbl3Cpb+IIcjBDUnnyBdQbq4w==",
-      "license": "MIT"
+      "license": "MIT",
+      "peer": true
    },
    "node_modules/redux-thunk": {
      "version": "3.1.0",
@@ -8966,6 +8980,7 @@
      "integrity": "sha512-kgW13M54DUB7IsIRM5LvJkNlpH+WhMpooUcaWGFARkF1Tc82v9mIWkCbCYf+MBvpIUBSeSOTilpZjEPr2VYE6Q==",
      "dev": true,
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "chokidar": "^4.0.0",
        "immutable": "^5.1.5",
@@ -10069,6 +10084,7 @@
      "integrity": "sha512-wGN3qcrBQIFmQ/c0AiOAQBvrZ5lmY8vbbMv4Mxfgzqd/B6+9pXtLo73WuS1dSGXM5QYY3hZnIbvx+K1xxe6FyA==",
      "dev": true,
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "@types/eslint-scope": "^3.7.7",
        "@types/estree": "^1.0.8",
@@ -10117,6 +10133,7 @@
      "integrity": "sha512-pIDJHIEI9LR0yxHXQ+Qh95k2EvXpWzZ5l+d+jIo+RdSm9MiHfzazIxwwni/p7+x4eJZuvG1AJwgC4TNQ7NRgsg==",
      "dev": true,
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "@discoveryjs/json-ext": "^0.5.0",
        "@webpack-cli/configtest": "^2.1.1",
@@ -1034,42 +1034,17 @@ const AgentChat: React.FC<AgentChatProps> = ({ sessionId: sessionIdProp, onClose
                    );
                  })()}
                  {(() => {
-                    const liveWindow = session.context_window || contextEstimate.limit || 200_000;
-                    const liveInput = session.tokens?.input ?? 0;
-                    const pct = liveInput > 0
-                      ? Math.min(1, liveInput / Math.max(1, liveWindow))
-                      : (contextEstimate.used / Math.max(1, liveWindow));
                    const mcpCount = session.active_mcps?.length ?? 0;
-                    // Quiet until it matters: hide the memory meter until the chat is filling up,
-                    // and hide the tool count unless tools are actually connected.
-                    const showMemory = pct >= 0.60;
-                    const showTools = mcpCount > 0;
-                    if (!showMemory && !showTools) return null;
-                    const pctTxt = `${Math.round(pct * 100)}%`;
-                    const memColor = pct >= 0.85 ? '#ef4444' : '#f59e0b';
-                    const tip = [
-                      showMemory ? `Memory ${pctTxt} full. As the chat fills up, the oldest messages start dropping out.` : null,
-                      showTools ? `${mcpCount} tool${mcpCount === 1 ? '' : 's'} connected.` : null,
-                    ].filter(Boolean).join('\n');
+                    if (mcpCount === 0) return null;
                    return (
                      <Typography
                        variant="caption"
                        sx={{ display: 'inline-flex', alignItems: 'center', gap: 0.5, fontVariantNumeric: 'tabular-nums' }}
-                        title={tip}
+                        title={`${mcpCount} tool${mcpCount === 1 ? '' : 's'} connected.`}
                      >
-                        {showMemory && (
-                          <Box component="span" sx={{ color: memColor, fontWeight: 500 }}>
-                            Memory {pctTxt} full
-                          </Box>
-                        )}
-                        {showMemory && showTools && (
-                          <Box component="span" sx={{ color: c.text.ghost }}>·</Box>
-                        )}
-                        {showTools && (
-                          <Box component="span" sx={{ color: c.text.tertiary }}>
-                            {mcpCount} tool{mcpCount === 1 ? '' : 's'}
-                          </Box>
-                        )}
+                        <Box component="span" sx={{ color: c.text.tertiary }}>
+                          {mcpCount} tool{mcpCount === 1 ? '' : 's'}
+                        </Box>
                      </Typography>
                    );
                  })()}
@@ -809,9 +809,6 @@ const AgentCard: React.FC<Props> = ({
          <Typography variant="caption" sx={{ color: c.text.tertiary }}>
            {friendlyModelLabel}
          </Typography>
-          <Typography variant="caption" sx={{ color: c.text.tertiary }}>
-            {session.mode}
-          </Typography>
          <Typography variant="caption" sx={{ color: c.text.tertiary }}>
            <ElapsedTimer messages={session.messages} status={session.status} />
          </Typography>