Tools
Loki S Insight
A lightweight visual dashboard to inspect and edit OpenClaw AI agent memory and system prompts.
Install
npm install
npm
README
# Loki's Insight 🔬
> **What is your agent actually thinking?**
Loki's Insight is a local web tool for inspecting and editing the context injected into [OpenClaw](https://github.com/openclaw/openclaw) agent sessions. Instead of treating your agent as a black box, you can see exactly what memory files were loaded, what was actually injected at runtime, and fix errors directly.
## Features
- **Static view** — Shows all workspace files loaded at session startup (SOUL.md, USER.md, MEMORY.md, daily logs). Edit and save directly in the browser.
- **Dynamic view** — Parses the actual session JSONL to show what was actually injected at runtime, including file reads, user messages, and agent outputs.
- **Skills browser** — Browse all installed skills (user / system / extension) and read their SKILL.md directly.
- **In-browser editor** — Edit `MEMORY.md` and other workspace files without touching the terminal.
- **i18n** — UI available in 中文 / English / 日本語 / 한국어.
## Why
OpenClaw loads memory files into the context window at session start. Over time these files can contain outdated info, errors, or redundant content — but there's no built-in way to review them. Loki's Insight fills that gap.
## Requirements
- Node.js 18+
- OpenClaw installed and running
## Installation
```bash
git clone https://github.com/Uplay111/Loki-s-Insight-
cd Loki-s-Insight-
npm install
npm run dev
```
Then open http://localhost:19999 in your browser.
### Platform notes
**macOS / Linux** — Works out of the box. OpenClaw is expected at `~/.openclaw`.
**Windows** — Works out of the box. OpenClaw is expected at `%USERPROFILE%\.openclaw`.
If your OpenClaw data is in a different location, set the `OPENCLAW_DIR` environment variable:
```powershell
# PowerShell
$env:OPENCLAW_DIR = "C:\path\to\your\openclaw"
npm run dev
```
```bash
# macOS / Linux
OPENCLAW_DIR=/custom/path npm run dev
```
## How it works
Loki's Insight reads data directly from the filesystem without modifying OpenClaw:
1. **`~/.openclaw/workspace/*.md`** — Static files loaded at session startup
2. **`~/.openclaw/agents/main/sessions/*.jsonl`** — Session transcripts containing `toolResult` entries that show what the agent actually read
3. **`~/.openclaw/workspace/skills/` and OpenClaw npm package** — Installed skills
The server is built with Hono + Node.js and serves a single-page UI.
## Roadmap
- [ ] Memory timeline — track how MEMORY.md evolved across sessions
- [ ] Token count estimation per block
- [ ] File read failure highlighting
- [ ] Watch mode (auto-refresh on file changes)
- [ ] Support for multi-agent workspaces
## License
MIT
tools
Comments
Sign in to leave a comment