fix: inject SessionDB into AIAgent for WebUI sessions (enables session_search)

[DelightRun]

Problem

The session_search tool always returns "Session database not available" when used from WebUI sessions. This means the agent cannot recall past conversations or search session history — a capability that works fine in CLI and gateway modes.

Root Cause

The session_search tool requires a SessionDB instance passed via the session_db parameter on AIAgent.init(). Both the CLI path (cli.py) and gateway path (gateway/run.py) already initialize and inject SessionDB, but the WebUI streaming path (api/streaming.py) was missing this step.

The agent's _invoke_tool() method checks self._session_db — when it's None, it short-circuits with the error message without ever touching the database.

Fix

Two changes in api/streaming.py:

1. Initialize SessionDB() before creating the AIAgent instance (with try/except so failures are non-fatal — just logs a warning)
2. Pass session_db=_session_db to the AIAgent() constructor

Testing

Verified manually: after the fix, session_search successfully returns recent sessions with metadata (session_id, title, source, timestamps, preview) from the existing state.db (59 sessions, 2309 messages).

[nesquena-hermes]

Review: fix — inject SessionDB into AIAgent for WebUI sessions

Thanks for this! Clean, minimal fix that closes an important capability gap.

What the fix does

session_search was silently returning "Session database not available" from WebUI sessions because api/streaming.py never initialized or passed a SessionDB instance — even though the CLI and gateway paths already do this. The two-line fix adds the missing initialization and passes it through.

Code review

The change is correct and well-structured:

_session_db = None
try:
    from hermes_state import SessionDB
    _session_db = SessionDB()
except Exception as _db_err:
    print(f"[webui] WARNING: SessionDB init failed — session_search will be unavailable: {_db_err}", flush=True)

• ✅ Non-fatal fallback — if hermes_state isn't available or SessionDB() raises, the agent still starts and session_search degrades gracefully (same behavior as before this PR)
• ✅ The flush=True on the warning ensures it shows up immediately in streamed log output
• ✅ session_db=_session_db passed at the correct position in the constructor call — matches the existing CLI/gateway pattern exactly
• ✅ Scoped inside the streaming generator where _AIAgent is resolved, not at module level — so it follows the same lazy-init pattern as the rest of streaming.py

Minor note

The warning uses print() rather than logger.debug(). Given that PR #354 (currently open) is replacing print/bare-except patterns with logger.debug() throughout the codebase, you may want to align this with that style — e.g.:

logger.warning("SessionDB init failed — session_search will be unavailable: %s", _db_err)

Not a blocker for this PR, just worth noting for consistency.

Summary

Check	Status
Root cause analysis	✅ Correct — CLI/gateway inject SessionDB, WebUI didn't
Fix correctness	✅ Minimal, targeted, non-breaking
Graceful degradation	✅ None fallback preserves prior behavior on failure
Scope	✅ Single file, 8 lines

Looks good to merge.

[nesquena-hermes]

Closed in favor of #359 which merged this fix with added tests (test_sprint42.py) and CHANGELOG entry. 822 tests pass on master.