An AI agent with a persistent knowledge graph as its identity — not static prompts, not stateless chat. Every conversation adds nodes, edges, and reflections. Over time, it forms a self.
This setup takes a minute. You'll pick a look, create an account, and connect at least one model provider.
Pick a theme
Set the look. You can change this any time in Settings.
Introduce yourself
How should the agent refer to you and be referred to?
The agent will answer to any of these.
Create your account
Secure the instance with a username + password. You'll sign in with this next time.
Connect a provider
Add at least one API key so the agent can run. Test each to verify it works.
Custom OAI-compatible providers
Add as many self-hosted or third-party OAI endpoints as you want. Each gets a unique prefix (e.g. glm, qwen) usable in the model tier step.
Route model tiers
Pick a provider + model for each tier. Planner is required; others can mirror it.
Vision & audio models (optional)
Specialist tiers for image / video / audio understanding. Skip to use the planner tier for everything.
Voice & web search (optional)
Turn on voice I/O and pick a web search provider. All optional.
Voice providers use the API keys you entered in step 4.
SearXNG is tried first. Brave is used when SearXNG is unset or empty.
What should the agent call you?
A short name or nickname. Defaults to your username.
You're all set 🎉
Your account is ready. Click Continue to start chatting with the agent and exploring the graph.
Browser backend
Which browser engine should tools use by default?
You're all set. Click Finish to save and start using your agent.
disconnected
thinking...
listening...
◀ Sidebar
Graph
Chat ▶
○ Graph
{ } Code ×
Select a node to inspect
People
Projects
{ }
Settings
Graphs
Theme
LongMemEval
Enhanced RecallOFF
LongMemEval
/500 (stratified across types)
~$3-5 • ~30-45 min (oracle)
STARTING
00:00
0 / 0
0
Sessions
0
Nodes
0
Answered
--
Accuracy
Category
Pass
Acc
--
Avg ms/q
--
Input tok
--
Output tok
--
Est. cost
Type0 nodes
no nodes
Terminallocal shell
SSH Hosts
🔒 Keystore locked — enter a passphrase to encrypt/decrypt SSH keys
This passphrase is held in memory only — never stored. You'll need it again after restart.
🔓 Keystore unlocked
Add Host
Node
Select a node on the graph
—
—
Files
idle
⚠ Sync is active — keep this tab open. Files at
Syncing...
workspace
Drop files here or click Upload
Logs
[click Refresh to load logs]
Skills
New Skill
New Knowledge Graph
Mount Local Folder
Settings
Appearance
Theme is saved per user. The sections below are agent-level settings.
Knowledge Graph
Switch the active knowledge graph for this agent. Each graph has its own nodes, edges, and memory.
Tools
Launches the LongMemEval HUD. Requires Enhanced Recall for best results.
Forces an immediate maintainer cycle (bypasses the 15-min interval gate). Triggers gap detection/filling, reflections, stale check, sparse-node connection, dedup, reasoning, and embedding passes.
Identity
Users
Manage who can access this instance. Changes take effect immediately and active sessions are dropped.
Your Profile
Change password
Memory
Uses LLM-assisted query decomposition for better temporal and cross-session retrieval.
Proactive
Voice
Runtime
Set this so the agent knows its own external address for generating links (web_serve, graph viewer, etc.). No trailing slash.
Agent ID
Web Port
Workspace
Data Dir
Model Routing
Casual
Normal
Planner
Subagent
Learner
Image VLM
Video VLM
Audio VLM
Providers
Anthropic / Claude
OpenAI
OpenRouter
Local / OAI-Compatible
Gemini Embedder
Used for node/attribute embeddings only. Not called for text generation.
Custom OAI-Compatible Providers
Acorn
Tick to enable; on save a key is auto-generated and shown here. Existing CLI users will need the new key if you regen.
Browser
Web Search
SearXNG (primary)
Self-hosted metasearch. Tried first; leave empty to disable.