personal memory agent
feat: entity enrichment observations from Granola importer
Extend seed_entities() to accept optional observations list per entity
dict. After entity creation/resolution, adds observations via
add_observation() with dedup against existing content.
Update Granola importer to build observation strings from parsed
participant data (title, company, LinkedIn) and pass them through
seed_entities(). Format: "fact (via Granola, YYYY-MM-DD)".
Implements product decision: records/decisions/2026-03-14-entity-enrichment-from-imports.md
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+255
-6
+208
tests/test_importer_granola.py
+208
tests/test_importer_granola.py
···
587
587
assert "Total:" in captured.out
588
588
assert "Available to import:" in captured.out
589
589
assert "Q1 Planning" in captured.out # title shown in available list
590
+
591
+
592
+
# ---------------------------------------------------------------------------
593
+
# Entity enrichment observations
594
+
# ---------------------------------------------------------------------------
595
+
596
+
ENRICHED_TRANSCRIPT = dedent("""\
597
+
---
598
+
doc_id: doc_enriched_001
599
+
source: granola
600
+
created_at: "2025-10-28T15:04:05Z"
601
+
remote_updated_at: "2025-10-29T01:23:45Z"
602
+
title: Enriched Meeting
603
+
duration_seconds: 600
604
+
generator: muesli 1.0
605
+
---
606
+
607
+
# Enriched Meeting
608
+
609
+
## Participants
610
+
611
+
- **Alice Smith**, Engineering Manager, Acme Corp, (alice@acme.com)
612
+
- **Bob Jones**, (bob@jones.io)
613
+
- **Jane Doe**, CTO, StartupCo, linkedin.com/in/janedoe, (jane@startup.co)
614
+
- **Carlos Garcia**, Designer, (carlos@example.com)
615
+
- **Eve Wong**, (eve@megacorp.com)
616
+
617
+
---
618
+
619
+
**Alice Smith (15:04:12):** Welcome everyone.
620
+
**Bob Jones (15:04:19):** Thanks for setting this up.
621
+
**Jane Doe (15:04:35):** Happy to be here.
622
+
**Carlos Garcia (15:05:01):** Same here.
623
+
**Eve Wong (15:05:10):** Let's get started.
624
+
""")
625
+
626
+
627
+
def test_observations_created_on_import(tmp_path, monkeypatch):
628
+
"""Observations are created for participants with enrichment data."""
629
+
from think.entities.observations import load_observations
630
+
from think.importers.granola import GranolaBackend
631
+
632
+
monkeypatch.setenv("JOURNAL_PATH", str(tmp_path))
633
+
634
+
muesli_dir = tmp_path / "muesli"
635
+
_write_transcript(muesli_dir, "2025-10-28_enriched.md", ENRICHED_TRANSCRIPT)
636
+
637
+
GranolaBackend().sync(tmp_path, source_path=muesli_dir, dry_run=False)
638
+
639
+
# Alice: title + company
640
+
alice_obs = load_observations("import.granola", "Alice Smith")
641
+
alice_contents = [o["content"] for o in alice_obs]
642
+
assert "Engineering Manager at Acme Corp (via Granola, 2025-10-28)" in alice_contents
643
+
644
+
# Bob: no title, no company, no linkedin — no observations
645
+
bob_obs = load_observations("import.granola", "Bob Jones")
646
+
assert len(bob_obs) == 0
647
+
648
+
# Jane: title + company + linkedin
649
+
jane_obs = load_observations("import.granola", "Jane Doe")
650
+
jane_contents = [o["content"] for o in jane_obs]
651
+
assert "CTO at StartupCo (via Granola, 2025-10-28)" in jane_contents
652
+
assert "LinkedIn: linkedin.com/in/janedoe (via Granola, 2025-10-28)" in jane_contents
653
+
654
+
# Carlos: title only (Designer, no company)
655
+
carlos_obs = load_observations("import.granola", "Carlos Garcia")
656
+
carlos_contents = [o["content"] for o in carlos_obs]
657
+
assert "Designer (via Granola, 2025-10-28)" in carlos_contents
658
+
659
+
# Eve: no title, no company, no linkedin — no observations
660
+
eve_obs = load_observations("import.granola", "Eve Wong")
661
+
assert len(eve_obs) == 0
662
+
663
+
664
+
def test_observations_not_duplicated_on_reimport(tmp_path, monkeypatch):
665
+
"""Re-importing the same transcript does not duplicate observations."""
666
+
from think.entities.observations import load_observations
667
+
from think.importers.granola import GranolaBackend
668
+
669
+
monkeypatch.setenv("JOURNAL_PATH", str(tmp_path))
670
+
671
+
muesli_dir = tmp_path / "muesli"
672
+
_write_transcript(muesli_dir, "2025-10-28_enriched.md", ENRICHED_TRANSCRIPT)
673
+
674
+
backend = GranolaBackend()
675
+
676
+
# First import
677
+
backend.sync(tmp_path, source_path=muesli_dir, dry_run=False)
678
+
alice_obs_1 = load_observations("import.granola", "Alice Smith")
679
+
assert len(alice_obs_1) == 1
680
+
681
+
# Second import (force to re-import)
682
+
backend.sync(tmp_path, source_path=muesli_dir, dry_run=False, force=True)
683
+
alice_obs_2 = load_observations("import.granola", "Alice Smith")
684
+
assert len(alice_obs_2) == 1 # still just one, not duplicated
685
+
686
+
jane_obs = load_observations("import.granola", "Jane Doe")
687
+
assert len(jane_obs) == 2 # title+company and linkedin, still two
688
+
689
+
690
+
def test_observations_source_day(tmp_path, monkeypatch):
691
+
"""Observation source_day matches the segment day, not today."""
692
+
from think.entities.observations import load_observations
693
+
from think.importers.granola import GranolaBackend
694
+
695
+
monkeypatch.setenv("JOURNAL_PATH", str(tmp_path))
696
+
697
+
muesli_dir = tmp_path / "muesli"
698
+
_write_transcript(muesli_dir, "2025-10-28_enriched.md", ENRICHED_TRANSCRIPT)
699
+
700
+
GranolaBackend().sync(tmp_path, source_path=muesli_dir, dry_run=False)
701
+
702
+
alice_obs = load_observations("import.granola", "Alice Smith")
703
+
assert len(alice_obs) == 1
704
+
assert alice_obs[0]["source_day"] == "20251028"
705
+
706
+
707
+
def test_seed_entities_without_observations(tmp_path, monkeypatch):
708
+
"""seed_entities() works unchanged when no observations are provided."""
709
+
from think.importers.shared import seed_entities
710
+
711
+
monkeypatch.setenv("JOURNAL_PATH", str(tmp_path))
712
+
713
+
entities = [
714
+
{"name": "Test Person", "type": "Person", "email": "test@example.com"},
715
+
]
716
+
result = seed_entities("test.facet", "20251028", entities)
717
+
assert len(result) == 1
718
+
assert result[0]["name"] == "Test Person"
719
+
720
+
721
+
def test_seed_entities_observation_formatting(tmp_path, monkeypatch):
722
+
"""seed_entities() creates observations with correct formatting for all field combos."""
723
+
from think.entities.observations import load_observations
724
+
from think.importers.shared import seed_entities
725
+
726
+
monkeypatch.setenv("JOURNAL_PATH", str(tmp_path))
727
+
728
+
entities = [
729
+
# title + company
730
+
{
731
+
"name": "Person A",
732
+
"type": "Person",
733
+
"observations": ["VP Engineering at BigCo (via Granola, 2025-10-28)"],
734
+
},
735
+
# title only
736
+
{
737
+
"name": "Person B",
738
+
"type": "Person",
739
+
"observations": ["Designer (via Granola, 2025-10-28)"],
740
+
},
741
+
# company only
742
+
{
743
+
"name": "Person C",
744
+
"type": "Person",
745
+
"observations": ["Works at MegaCorp (via Granola, 2025-10-28)"],
746
+
},
747
+
# linkedin
748
+
{
749
+
"name": "Person D",
750
+
"type": "Person",
751
+
"observations": ["LinkedIn: linkedin.com/in/persond (via Granola, 2025-10-28)"],
752
+
},
753
+
]
754
+
seed_entities("test.facet", "20251028", entities)
755
+
756
+
a_obs = load_observations("test.facet", "Person A")
757
+
assert len(a_obs) == 1
758
+
assert a_obs[0]["content"] == "VP Engineering at BigCo (via Granola, 2025-10-28)"
759
+
assert a_obs[0]["source_day"] == "20251028"
760
+
761
+
b_obs = load_observations("test.facet", "Person B")
762
+
assert len(b_obs) == 1
763
+
assert b_obs[0]["content"] == "Designer (via Granola, 2025-10-28)"
764
+
765
+
c_obs = load_observations("test.facet", "Person C")
766
+
assert len(c_obs) == 1
767
+
assert c_obs[0]["content"] == "Works at MegaCorp (via Granola, 2025-10-28)"
768
+
769
+
d_obs = load_observations("test.facet", "Person D")
770
+
assert len(d_obs) == 1
771
+
assert d_obs[0]["content"] == "LinkedIn: linkedin.com/in/persond (via Granola, 2025-10-28)"
772
+
773
+
774
+
def test_seed_entities_observation_dedup(tmp_path, monkeypatch):
775
+
"""seed_entities() does not duplicate observations on re-call."""
776
+
from think.entities.observations import load_observations
777
+
from think.importers.shared import seed_entities
778
+
779
+
monkeypatch.setenv("JOURNAL_PATH", str(tmp_path))
780
+
781
+
entities = [
782
+
{
783
+
"name": "Dedup Person",
784
+
"type": "Person",
785
+
"observations": ["CTO at Acme (via Granola, 2025-10-28)"],
786
+
},
787
+
]
788
+
789
+
# First call creates the observation
790
+
seed_entities("test.facet", "20251028", entities)
791
+
obs = load_observations("test.facet", "Dedup Person")
792
+
assert len(obs) == 1
793
+
794
+
# Second call with same observation does not duplicate
795
+
seed_entities("test.facet", "20251028", entities)
796
+
obs = load_observations("test.facet", "Dedup Person")
797
+
assert len(obs) == 1
+32
-5
think/importers/granola.py
+32
-5
think/importers/granola.py
···
261
261
participants = _parse_participants(body)
262
262
if participants and segments:
263
263
first_day = segments[0][0]
264
-
entity_dicts = [
265
-
{"name": p["name"], "type": "Person", "email": p["email"]}
266
-
for p in participants
267
-
if p["name"]
268
-
]
264
+
# Format day for observation attribution: YYYYMMDD -> YYYY-MM-DD
265
+
obs_date = f"{first_day[:4]}-{first_day[4:6]}-{first_day[6:8]}"
266
+
entity_dicts = []
267
+
for p in participants:
268
+
if not p["name"]:
269
+
continue
270
+
d: dict[str, Any] = {
271
+
"name": p["name"],
272
+
"type": "Person",
273
+
"email": p["email"],
274
+
}
275
+
observations: list[str] = []
276
+
title = p.get("title", "")
277
+
company = p.get("company", "")
278
+
linkedin = p.get("linkedin", "")
279
+
if title and company:
280
+
observations.append(
281
+
f"{title} at {company} (via Granola, {obs_date})"
282
+
)
283
+
elif title:
284
+
observations.append(f"{title} (via Granola, {obs_date})")
285
+
elif company:
286
+
observations.append(
287
+
f"Works at {company} (via Granola, {obs_date})"
288
+
)
289
+
if linkedin:
290
+
observations.append(
291
+
f"LinkedIn: linkedin.com/in/{linkedin} (via Granola, {obs_date})"
292
+
)
293
+
if observations:
294
+
d["observations"] = observations
295
+
entity_dicts.append(d)
269
296
if entity_dicts:
270
297
try:
271
298
seeded = seed_entities("import.granola", first_day, entity_dicts)