Files
the-nexus/nexus/experience_store.py
Alexander Whitestone 0a0a2eb802
Some checks failed
CI / test (pull_request) Failing after 9s
CI / validate (pull_request) Failing after 15s
Review Approval Gate / verify-review (pull_request) Failing after 3s
fix: closes #1181
2026-04-12 12:18:55 -04:00

299 lines
11 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
"""
Nexus Experience Store — Embodied Memory
SQLite-backed store for lived experiences only. The model remembers
what it perceived, what it thought, and what it did — nothing else.
Each row is one cycle of the perceive→think→act loop.
Implements the GBrain "compiled truth + timeline" pattern (#1181):
- compiled_truths: current best understanding, rewritten when evidence changes
- experiences: append-only evidence trail that never gets edited
"""
import sqlite3
import json
import time
from pathlib import Path
from typing import Optional
DEFAULT_DB = Path.home() / ".nexus" / "experience.db"
MAX_CONTEXT_EXPERIENCES = 20 # Recent experiences fed to the model
class ExperienceStore:
def __init__(self, db_path: Optional[Path] = None):
self.db_path = db_path or DEFAULT_DB
self.db_path.parent.mkdir(parents=True, exist_ok=True)
self.conn = sqlite3.connect(str(self.db_path))
self.conn.execute("PRAGMA journal_mode=WAL")
self.conn.execute("PRAGMA synchronous=NORMAL")
self._init_tables()
def _init_tables(self):
self.conn.executescript("""
CREATE TABLE IF NOT EXISTS experiences (
id INTEGER PRIMARY KEY AUTOINCREMENT,
timestamp REAL NOT NULL,
perception TEXT NOT NULL,
thought TEXT,
action TEXT,
action_result TEXT,
cycle_ms INTEGER DEFAULT 0,
session_id TEXT
);
CREATE TABLE IF NOT EXISTS summaries (
id INTEGER PRIMARY KEY AUTOINCREMENT,
timestamp REAL NOT NULL,
summary TEXT NOT NULL,
exp_start INTEGER NOT NULL,
exp_end INTEGER NOT NULL
);
CREATE INDEX IF NOT EXISTS idx_exp_ts
ON experiences(timestamp DESC);
CREATE INDEX IF NOT EXISTS idx_exp_session
ON experiences(session_id);
-- GBrain compiled truth pattern (#1181)
-- Current best understanding about an entity/topic.
-- Rewritten when new evidence changes the picture.
-- The timeline (experiences table) is the evidence trail — never edited.
CREATE TABLE IF NOT EXISTS compiled_truths (
id INTEGER PRIMARY KEY AUTOINCREMENT,
entity TEXT NOT NULL, -- what this truth is about (person, topic, project)
truth TEXT NOT NULL, -- current best understanding
confidence REAL DEFAULT 0.5, -- 0.01.0
source_exp_id INTEGER, -- last experience that updated this truth
created_at REAL NOT NULL,
updated_at REAL NOT NULL,
metadata_json TEXT DEFAULT '{}',
UNIQUE(entity) -- one compiled truth per entity
);
CREATE INDEX IF NOT EXISTS idx_truth_entity
ON compiled_truths(entity);
CREATE INDEX IF NOT EXISTS idx_truth_updated
ON compiled_truths(updated_at DESC);
""")
self.conn.commit()
def record(
self,
perception: str,
thought: Optional[str] = None,
action: Optional[str] = None,
action_result: Optional[str] = None,
cycle_ms: int = 0,
session_id: Optional[str] = None,
) -> int:
"""Record one perceive→think→act cycle."""
cur = self.conn.execute(
"""INSERT INTO experiences
(timestamp, perception, thought, action, action_result,
cycle_ms, session_id)
VALUES (?, ?, ?, ?, ?, ?, ?)""",
(time.time(), perception, thought, action,
action_result, cycle_ms, session_id),
)
self.conn.commit()
return cur.lastrowid
def recent(self, limit: int = MAX_CONTEXT_EXPERIENCES) -> list[dict]:
"""Fetch the most recent experiences for context."""
rows = self.conn.execute(
"""SELECT id, timestamp, perception, thought, action,
action_result, cycle_ms
FROM experiences
ORDER BY timestamp DESC
LIMIT ?""",
(limit,),
).fetchall()
return [
{
"id": r[0],
"timestamp": r[1],
"perception": r[2],
"thought": r[3],
"action": r[4],
"action_result": r[5],
"cycle_ms": r[6],
}
for r in reversed(rows) # Chronological order
]
def format_for_context(self, limit: int = MAX_CONTEXT_EXPERIENCES) -> str:
"""Format recent experiences as natural language for the model."""
experiences = self.recent(limit)
if not experiences:
return "You have no memories yet. This is your first moment."
lines = []
for exp in experiences:
ago = time.time() - exp["timestamp"]
if ago < 60:
when = f"{int(ago)}s ago"
elif ago < 3600:
when = f"{int(ago / 60)}m ago"
else:
when = f"{int(ago / 3600)}h ago"
line = f"[{when}] You perceived: {exp['perception']}"
if exp["thought"]:
line += f"\n You thought: {exp['thought']}"
if exp["action"]:
line += f"\n You did: {exp['action']}"
if exp["action_result"]:
line += f"\n Result: {exp['action_result']}"
lines.append(line)
return "Your recent experiences:\n\n" + "\n\n".join(lines)
def count(self) -> int:
"""Total experiences recorded."""
return self.conn.execute(
"SELECT COUNT(*) FROM experiences"
).fetchone()[0]
def save_summary(self, summary: str, exp_start: int, exp_end: int):
"""Store a compressed summary of a range of experiences.
Used when context window fills — distill old memories."""
self.conn.execute(
"""INSERT INTO summaries (timestamp, summary, exp_start, exp_end)
VALUES (?, ?, ?, ?)""",
(time.time(), summary, exp_start, exp_end),
)
self.conn.commit()
def get_summaries(self, limit: int = 5) -> list[dict]:
"""Fetch recent experience summaries."""
rows = self.conn.execute(
"""SELECT id, timestamp, summary, exp_start, exp_end
FROM summaries ORDER BY timestamp DESC LIMIT ?""",
(limit,),
).fetchall()
return [
{"id": r[0], "timestamp": r[1], "summary": r[2],
"exp_start": r[3], "exp_end": r[4]}
for r in reversed(rows)
]
def close(self):
self.conn.close()
# ── GBrain compiled truth + timeline pattern (#1181) ────────────────
def upsert_compiled_truth(
self,
entity: str,
truth: str,
confidence: float = 0.5,
source_exp_id: Optional[int] = None,
metadata: Optional[dict] = None,
) -> int:
"""Create or update the compiled truth for an entity.
This is the 'compiled truth on top' from the GBrain pattern.
When new evidence changes our understanding, we rewrite this
record. The timeline (experiences table) preserves what led
here — it is never edited.
Args:
entity: What this truth is about (person, topic, project).
truth: Current best understanding.
confidence: 0.01.0 confidence score.
source_exp_id: Last experience ID that informed this truth.
metadata: Optional extra data as a dict.
Returns:
The row ID of the compiled truth.
"""
now = time.time()
meta_json = json.dumps(metadata) if metadata else "{}"
self.conn.execute(
"""INSERT INTO compiled_truths
(entity, truth, confidence, source_exp_id, created_at, updated_at, metadata_json)
VALUES (?, ?, ?, ?, ?, ?, ?)
ON CONFLICT(entity) DO UPDATE SET
truth = excluded.truth,
confidence = excluded.confidence,
source_exp_id = excluded.source_exp_id,
updated_at = excluded.updated_at,
metadata_json = excluded.metadata_json""",
(entity, truth, confidence, source_exp_id, now, now, meta_json),
)
self.conn.commit()
row = self.conn.execute(
"SELECT id FROM compiled_truths WHERE entity = ?", (entity,)
).fetchone()
return row[0]
def get_compiled_truth(self, entity: str) -> Optional[dict]:
"""Get the current compiled truth for an entity."""
row = self.conn.execute(
"""SELECT id, entity, truth, confidence, source_exp_id,
created_at, updated_at, metadata_json
FROM compiled_truths WHERE entity = ?""",
(entity,),
).fetchone()
if not row:
return None
return {
"id": row[0],
"entity": row[1],
"truth": row[2],
"confidence": row[3],
"source_exp_id": row[4],
"created_at": row[5],
"updated_at": row[6],
"metadata": json.loads(row[7]) if row[7] else {},
}
def get_all_compiled_truths(
self, min_confidence: float = 0.0, limit: int = 100
) -> list[dict]:
"""Get all compiled truths, optionally filtered by minimum confidence."""
rows = self.conn.execute(
"""SELECT id, entity, truth, confidence, source_exp_id,
created_at, updated_at, metadata_json
FROM compiled_truths
WHERE confidence >= ?
ORDER BY updated_at DESC
LIMIT ?""",
(min_confidence, limit),
).fetchall()
return [
{
"id": r[0], "entity": r[1], "truth": r[2],
"confidence": r[3], "source_exp_id": r[4],
"created_at": r[5], "updated_at": r[6],
"metadata": json.loads(r[7]) if r[7] else {},
}
for r in rows
]
def search_compiled_truths(self, query: str, limit: int = 10) -> list[dict]:
"""Search compiled truths by entity name or truth content (LIKE match)."""
rows = self.conn.execute(
"""SELECT id, entity, truth, confidence, source_exp_id,
created_at, updated_at, metadata_json
FROM compiled_truths
WHERE entity LIKE ? OR truth LIKE ?
ORDER BY confidence DESC, updated_at DESC
LIMIT ?""",
(f"%{query}%", f"%{query}%", limit),
).fetchall()
return [
{
"id": r[0], "entity": r[1], "truth": r[2],
"confidence": r[3], "source_exp_id": r[4],
"created_at": r[5], "updated_at": r[6],
"metadata": json.loads(r[7]) if r[7] else {},
}
for r in rows
]