返回目录
模型与 MCP 插件

perseus-vault-dsh

Perseus-Computing-LLC/perseus-vault-dsh

Perseus Vault integration for DeepSeek Harness: MCP config example + native pre-step memory-injection plugin

Stars
0
Forks
0
Issues
0
更新
今天

PROJECT TOPICS

项目标签

PROJECT README

README

perseus-vault-dsh

Perseus Vault integration for DeepSeek Harness (dsh) — the open-source agent harness where everything is a plugin.

Two ways to connect the vault to a dsh agent:

Mode What it gives you File
MCP config row (recommended start) The vault's agent-safe tool surface as mcp__perseus_vault__* tools: remember, recall, recall_when, semantic_search, context, capture, journal, state, links, timelines — 48 tools, zero code examples/dsh-mcp/perseus-vault.cordis.yml
Native plugin Automatic pre-step memory injection: relevant memories are pushed into the system prompt before every user turn (no "remember to call recall" needed) packages/perseus-vault

Both can be mounted together — the plugin handles injection, the MCP row gives the agent explicit memory tools.

Quickstart

# 1. Install the Perseus Vault binary (pinned release)
cargo install perseus-vault@2.23.0   # or grab a release binary / Docker image

# 2. Run dsh with the MCP overlay
npx @deepseek-ai/dsh web --patch "$PWD/examples/dsh-mcp/perseus-vault.cordis.yml"

The agent now sees mcp__perseus_vault__recall, mcp__perseus_vault__remember, etc. For automatic injection, also mount the plugin:

- insert:
    - id: perseus-vault-memory
      name: '@perseus-computing/dsh'   # or a local path while testing
      config:
        command: perseus-vault

See examples/dsh-mcp/README.md and packages/perseus-vault/README.md for details.

What is Perseus Vault?

A persistent, encrypted, tamper-evident memory engine for AI agents (MIT). Durable facts with decay, bitemporal history, contradiction detection, workspace scoping, and an audit journal — one brain across projects and clients. Perseus-Computing-LLC/perseus-vault

License

MIT — see LICENSE.

CLASSIFICATION EVIDENCE

分类依据

项目类型插件
功能分类模型与 MCP
规则置信度

系统优先读取 GitHub Topics,再与站内分类词典和词根规则比对。