id: tencentdb_agent_memory_pinned
framework: tencentdb-agent-memory
mode: white_box
status: implemented
description: TencentDB Agent Memory target using the local standalone HTTP gateway, auto-capture, L1 extraction, recall, and SQLite evidence inspection.
source_url: https://github.com/TencentCloud/TencentDB-Agent-Memory
source:
  type: node_package_and_local_gateway
  repo: TencentCloud/TencentDB-Agent-Memory
  package: "@tencentdb-agent-memory/memory-tencentdb"
  version: 0.3.6
runtime:
  type: local_node_gateway
  package_env: TENCENTDB_AGENT_MEMORY_DIR
  required_env:
    - OPENAI_API_KEY
  llm:
    base_url: https://api.openai.com/v1
    model: gpt-4o-mini
    max_tokens: 2048
  memory:
    recall_strategy: keyword
    every_n_conversations: 1
    l1_idle_timeout_seconds: 1
model:
  provider: openai_compatible
  model: gpt-4o-mini
  temperature: 0
adapter:
  module: memorybench.adapters.tencentdb_agent_memory
  class: TencentDbAgentMemoryAdapter
capabilities:
  inspect_memory: true
  delete_memory: false
  retrieval_log: true
  multi_user: true
  multi_tenant: false
notes:
  - The adapter starts a fresh standalone TencentDB Agent Memory gateway process and data directory per benchmark subject to avoid cross-scenario contamination.
  - It uses the public /recall, /capture, /search/memories, /search/conversations, and /session/end HTTP routes. The public gateway does not expose direct record deletion, so deletion behavior is tested through the user's natural-language forget request and the resulting memory state.
  - Set TENCENTDB_AGENT_MEMORY_DIR to a checkout or installed package directory containing src/gateway/server.ts. The local run used the upstream package with Node.js 22.