1AIVault Documentation
Auto-Inject Memory
Write a managed memory brief into supported AI instruction files and Claude Code session hooks.
Auto-Inject Memory gives supported AI tools a compact working brief before they call MCP. It is useful for preferences, decisions, pinned memory, recent work, and known gaps.
Master Switch
The feature is opt-in. When the master switch is off, 1AIVault does not write managed memory blocks into instruction files.
Profiles
| Profile | Approximate budget | Notes |
|---|---|---|
| Minimal | 600 tokens | Critical saved context only |
| Standard | 1,800 tokens | Balanced default |
| Rich | 3,500 tokens | Pro profile |
| Max | 8,000 tokens | Pro profile |
Project Scope
Static file blocks can be global or filtered by project:* tags. The Claude Code hook can read the vault live at session start and is project-aware from the current working directory.
Targets
Targets include Claude Code, Codex, Gemini, Antigravity, Qwen, OpenCode, and Windsurf instruction files where supported. The UI shows whether a target is injected, managed by a hook, missing, or disabled.
Preview Before Writing
Use Preview to inspect the exact managed block before enabling or refreshing. This is the safest way to confirm that only the intended memory is being injected.
Test Prompt
After enabling a target, start a fresh AI session and ask it to audit the context already injected without calling tools. If it can summarize current projects, decisions, and gaps, the static brief is working.