Claude Code Hooks

Auto-inject memory context into Claude Code sessions

Hooks let Claude Code automatically retrieve your conversation history context, no manual search needed.

Default: Disabled. Hooks are off by default. You need to enable them in both Claude Code settings and Memex config.

How it works:

SessionStart - Injects recent conversation summaries when you open a new session
UserPromptSubmit - Searches relevant history based on your message before each request

Result: Claude "remembers" what you've been working on without you explaining the background every time.

Setup

Step 1: Configure Claude Code

Edit ~/.claude/settings.json:

{
  "hooks": {
    "SessionStart": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "curl -s -X POST http://localhost:10013/api/inject -H 'Content-Type: application/json' -d '{\"hook\":\"SessionStart\"}' 2>/dev/null || echo '{}'"
          }
        ]
      }
    ],
    "UserPromptSubmit": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "curl -s -X POST http://localhost:10013/api/inject -H 'Content-Type: application/json' -d '{\"hook\":\"UserPromptSubmit\",\"query\":\"$PROMPT\"}' 2>/dev/null || echo '{}'"
          }
        ]
      }
    ]
  }
}

Step 2: Enable injection in Memex

Edit ~/.vimo/memex/config.json:

{
  "compact": {
    "inject": {
      "mode": "full"
    }
  }
}

Step 3: Restart Claude Code

Done! Claude will now automatically retrieve history context.

Hook Types

Hook	Trigger	Effect
`SessionStart`	New session opened	Injects recent 10 conversation summaries
`UserPromptSubmit`	Before sending message	Searches history related to your question

SessionStart is good for: Letting Claude know what projects you've been working on.

UserPromptSubmit is good for: When you ask "how did I fix that bug before", it automatically finds relevant conversations.

Configuration

Quick Mode

{
  "compact": {
    "inject": {
      "mode": "full"
    }
  }
}

Mode	Description
`none`	Disabled (default)
`full`	Enable both hooks

Detailed Configuration

{
  "compact": {
    "inject": {
      "session_start": {
        "enabled": true,
        "max_items": 10,
        "max_tokens": 2000
      },
      "user_prompt": {
        "enabled": true,
        "similarity_threshold": 0.3,
        "max_tokens": 2000
      }
    }
  }
}

Option	Default	Description
`enabled`	`false`	Enable this hook
`max_items`	`10`	Number of items to inject (SessionStart)
`max_tokens`	`2000`	Maximum context tokens
`similarity_threshold`	`0.3`	Similarity threshold for search (lower = more results)

Requirements

Memex server running
SessionStart: No special requirements
UserPromptSubmit: Requires Ollama + Compact feature

Verify

# Test SessionStart
curl -X POST http://localhost:10013/api/inject \
  -H 'Content-Type: application/json' \
  -d '{"hook":"SessionStart"}'

# Test UserPromptSubmit
curl -X POST http://localhost:10013/api/inject \
  -H 'Content-Type: application/json' \
  -d '{"hook":"UserPromptSubmit","query":"test"}'

If hookSpecificOutput.additionalContext contains content, the setup is successful.