win-memory

name	win-memory
description	Persistent agent memory via OpenViking — store and retrieve context across sessions using a tiered filesystem database. L0/L1/L2 context layers with semantic search.
metadata	{"openclaw":{"emoji":"🧬","os":["win32"],"requires":{"bins":["python","openviking-server"]},"install":"pip install openviking"}}

win-memory

Persistent agent memory for OpenClaw on Windows, powered by OpenViking — an open-source context database designed for AI agents.

OpenViking stores memories, resources, and skills in a virtual filesystem (viking://) with three context tiers:

• L0 (Abstract, ~100 tokens) — quick relevance check
• L1 (Overview, ~2k tokens) — key information for planning
• L2 (Full content) — loaded on demand only

This means your agent loads context efficiently, not all at once.

Setup

1. Install OpenViking

winget install Python.Python.3.12
pip install openviking

2. Configure

$configDir = "$env:USERPROFILE\.openviking"
New-Item -ItemType Directory -Force -Path $configDir | Out-Null

# Minimal config using local Ollama (no API cost)
$config = @{
    storage = @{ workspace = "$env:USERPROFILE\.openviking\workspace" }
    log = @{ level = "INFO"; output = "stdout" }
    embedding = @{
        dense = @{
            api_base = "http://localhost:11434/v1"
            api_key = "ollama"
            provider = "openai"
            dimension = 768
            model = "nomic-embed-text"
        }
    }
    vlm = @{
        api_base = "http://localhost:11434/v1"
        api_key = "ollama"
        provider = "openai"
        model = "llama3.1"
    }
} | ConvertTo-Json -Depth 5
$config | Out-File -Encoding utf8 "$configDir\ov.conf"

Write-Host "OpenViking configured at $configDir\ov.conf"