Rich CLI progress not visible when output captured by non-TTY (subprocess/pipe)

[bm-clawd]

Problem

bm reindex --embeddings uses Rich progress bars that render correctly in a real terminal but are not visible when output is captured by a subprocess or piped (non-TTY context). This affects tools/plugins that shell out to bm CLI commands.

Observed Behavior

When called from a plugin subprocess:

Project: claw
  Building vector embeddings...
[... 74 seconds of silence ...]
    Embedding entities... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100%

The progress bar only appears at completion because Rich detects non-TTY and doesn't flush incremental updates.

Expected Behavior

When not connected to a TTY, fall back to simple log-style progress lines:

Embedded 10/66 entities...
Embedded 20/66 entities...

Impact

Low for direct CLI users. Confusing for plugin/tooling integrations — looks like a hang on slower hardware (74s on Intel Mac at 394% CPU).

Note

This is a known Rich library behavior. Fix options:

• Detect non-TTY and use Console(force_terminal=False) with plain text fallback
• Add --no-rich / --plain flag
• Log progress lines alongside Rich output

[claude]

Claude finished @bm-clawd's task in 1m 49s —— View job

---

Triage Analysis

• Analyze issue and review relevant code
• Check for duplicates
• Classify issue and add labels
• Provide triage summary

---

Issue Classification

Attribute	Assessment
Type	Enhancement (UX improvement)
Priority	Low
Component	CLI
Complexity	Simple

Labels added: enhancement

---

Root Cause Analysis

I reviewed the relevant code in src/basic_memory/cli/commands/db.py (lines 197-210) and src/basic_memory/services/search_service.py.

The progress bar setup looks correct at first glance:

with Progress(..., console=console) as progress:
    task = progress.add_task("  Embedding entities...", total=None)

    def on_progress(entity_id, index, total):
        progress.update(task, total=total, completed=index)

    stats = await search_service.reindex_vectors(progress_callback=on_progress)

The root issue is that embed_documents() blocks the event loop for the entire duration of each entity's embedding computation. On a local FastEmbed/CPU setup, this can be several seconds per entity. Since Rich's progress bar requires event loop cycles to re-render, the display never gets a chance to update between entities — the bar only appears to render once everything completes.

Additionally, for non-TTY capture (piped output), Rich buffers output rather than streaming it incrementally.

---

Suggested Fix

Two complementary approaches:

1. Force Rich to flush after each update — add console.file.flush() or use transient=False with explicit refresh() calls after each progress.update()
2. Run CPU-bound embedding in a thread pool — wrap embed_documents() with asyncio.to_thread() so the event loop remains free for Rich to re-render between entities

The on_progress callback fires before sync_entity_vectors() is awaited (i.e., progress_callback(entity.id, i, len(entities)) is called at the start of each iteration), which means the update is fired but the event loop never yields back before the next blocking call.

A minimal fix would be to add await asyncio.sleep(0) after each progress update in reindex_vectors, or to explicitly call progress.refresh() — but the more robust fix is offloading the blocking embedding call to a thread pool executor.

---

No duplicates found for this specific issue.