Dashboard
Activity (7 days)
Live Feed
→ Get started
Memories
Memories are stored automatically by connected AI tools or any app using the SDK, and retrieved by semantic similarity. Search using natural language — try "decisions about authentication" or "recent bugs".
Try broader terms, synonyms, or browse all memories.
Connect an AI tool or write to the vault via the SDK and memories appear here automatically.
→ Get started
Memory pairs flagged as potentially contradictory by the Contradiction Detection worker. Click View to inspect each memory in detail.
The contradiction worker will flag pairs that conflict with each other.
Graph
The memory graph shows how your memories are connected. Each node is an engram (memory); edges represent associations — supports, contradicts, depends on, etc. Click a node to preview its content.
The entity graph shows named entities (people, organizations, technologies, etc.) and their typed relationships. Each node represents an entity; edges show relationship types extracted from memories.
Observability
System Health
Storage (Pebble)
Processors
Cognitive Workers
Per-Vault Breakdown
| Op | p50 | p95 | p99 | avg | count |
|---|---|---|---|---|---|
Settings
Manage admin credentials. The default credentials (root / password) should be changed after first login.
Change Admin Password
Database Backup
Create a point-in-time backup of the database using Pebble's checkpoint.
Safe to run on a live server. The backup is written to ./backups/ relative to the server working directory.
API keys let AI agents access MuninnDB without admin credentials. Keys are scoped to a vault and a mode (full or observe).
Create New Key
Token (shown once — copy it now)
No API keys yet. Create one above.
| Label | Vault | Mode | Expires | Action |
|---|---|---|---|---|
Configuration and cognitive engine status for vault:
Vault Configuration
Semantic Search
Active InactiveSemantic search is disabled. Memories are found by keyword only.
Configure embedder →Cognitive Engine Workers
Background workers that apply Hebbian learning, temporal scoring, contradiction detection, and confidence updates. Configured via server startup flags.
| Worker | Status | Processed |
|---|---|---|
Write-time Processing
Plasticity
Controls how memories behave in this vault — temporal scoring, associative learning, and retrieval weights.
Choose a Preset
Default Recall Mode
Override individual preset values. Leave blank to use preset default.
Preset Comparison
Vault Actions
Destructive operations on vault . These cannot be undone.
Semantic search uses the built-in model by default. Upgrade to a cloud provider for higher-quality embeddings, or add LLM enrichment to auto-extract concepts from stored memories. Changes require a server restart.
ollama pull nomic-embed-text), then click Refresh above.ollama pull llama3.2), then click Refresh above.Connect AI tools via MCP or integrate any application using the SDK. Once connected, memories are stored and retrieved automatically.
MCP Endpoint
~/.muninn/mcp.token. The configs below include the required header.
⚠ No MCP token configured — endpoint is open. Run muninn init to add a token.
Claude Code / Claude CLI
Add via one command — open a terminal and run:
claude mcp add muninn
claude mcp add muninn \ --header "Authorization: Bearer $(cat ~/.muninn/mcp.token)"
Or auto-configure with:
muninn init --tool claude-code --yes
Config written to ~/.claude.json — no restart required.
Claude Desktop
Run muninn init --tool claude --yes to configure automatically, or add manually:
~/Library/Application Support/Claude/claude_desktop_config.json
{
"mcpServers": {
"muninn": {
"url": ""
}
}
}
{
"mcpServers": {
"muninn": {
"url": "",
"headers": {
"Authorization": "Bearer <token from ~/.muninn/mcp.token>"
}
}
}
}
Cursor
Run muninn init --tool cursor --yes or edit ~/.cursor/mcp.json:
{
"mcpServers": {
"muninn": {
"url": ""
}
}
}
{
"mcpServers": {
"muninn": {
"url": "",
"headers": {
"Authorization": "Bearer <token from ~/.muninn/mcp.token>"
}
}
}
}
VS Code
Create .vscode/mcp.json in your workspace:
{
"servers": {
"muninn": {
"url": ""
}
}
}
{
"servers": {
"muninn": {
"url": "",
"headers": {
"Authorization": "Bearer <token from ~/.muninn/mcp.token>"
}
}
}
}
Codex
Run muninn init --tool codex --yes or edit ~/.codex/config.toml:
[mcp_servers.muninn] url = ""
[mcp_servers.muninn] url = "" [mcp_servers.muninn.http_headers] Authorization = "Bearer <token from ~/.muninn/mcp.token>"
Auto-configure from CLI
The easiest way — handles token auth automatically:
muninn init
Session Timeline
Memories created or accessed by connected tools and applications in the selected time window.
Try a longer range, or get started to start building memory.
Server Logs
Cluster
Enable Cluster Mode
Connect this node to a MuninnDB cluster for high availability and replication.
⚠ Without a secret, any node can join this cluster
Enable Cluster Mode?
This will restart the replication subsystem. Existing data is not affected.
Cluster Members
| Node | Role | Status | Epoch | Replication Lag |
|---|---|---|---|---|
Auto-refreshes every 5 seconds
Cognitive Consistency
Auto-refreshes every 30 seconds
Failover Event Timeline
Security Posture
Live Activity
Join Token
Valid for minutes
TLS Certificate
mTLS
TLS 1.3, ECDSA P-256
Cluster Behavior
Heartbeat Interval
Milliseconds between PING messages
SDOWN Threshold
Missed heartbeats before marking SDOWN
CCS Probe Interval
Seconds between consistency checks
Reconcile on Heal
Run reconciliation after partition heals
Add Node
Remove Node
Remove from the cluster?
Trigger Failover
Regenerate Join Token?
The old token will immediately become invalid.
