47bd3ded44
Add an optional PostgreSQL backend (QMD_BACKEND=pg) alongside the unchanged default SQLite path. PG store uses pgvector (HNSW) for vectors and pg_jieba + pg_trgm for full-text/Chinese tokenization, with a namespace column isolating multi-agent memory (openclaw/hermes). - src/pg/: config, db-pg, schema bootstrap, memory store - MCP memory_add/memory_search/memory_get tools; qmd pg status + memory CLI - connection via QMD_PG_URL/DATABASE_URL/qmd config, stunnel TLS 5443 - tests: pg-config (unit) + pg-memory integration (gated on QMD_PG_URL) + pg-compose - docs/plan: plan, usage, test report, changelog; track docs/**/*.md SQLite path: zero regression (typecheck clean, 249 passed / 6 skipped). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
25 lines
836 B
YAML
25 lines
836 B
YAML
# Local PostgreSQL for qmd memory-bridge integration tests.
|
|
#
|
|
# docker compose -f test/pg-compose.yml up -d
|
|
# QMD_PG_URL='postgres://postgres:postgres@localhost:5432/postgres' \
|
|
# npx vitest run test/pg-memory.integration.test.ts
|
|
# docker compose -f test/pg-compose.yml down -v
|
|
#
|
|
# pgvector/pgvector ships pgvector but NOT pg_jieba, so FTS falls back to the
|
|
# 'english' text-search config. For full Chinese tokenization use the
|
|
# postgresql.svc.plus image (ghcr.io/x-evor/images/postgresql) which bundles
|
|
# pgvector + pg_jieba + pg_trgm.
|
|
services:
|
|
qmd-pg:
|
|
image: pgvector/pgvector:pg16
|
|
environment:
|
|
POSTGRES_PASSWORD: postgres
|
|
POSTGRES_DB: postgres
|
|
ports:
|
|
- "5432:5432"
|
|
healthcheck:
|
|
test: ["CMD-SHELL", "pg_isready -U postgres"]
|
|
interval: 2s
|
|
timeout: 5s
|
|
retries: 20
|