Connecting to /context…
Live recall
Watch the memories that shape each incoming request.
Following latest
Iframe graph
Connecting…
Waiting for /context
No context request yet
Keep this tab open while your client chats with the character.
Activation field
No graph in this request
Knowledge graph idle
When graph memory is recalled, its activation cloud will appear here.
self
person
fact
episode
entity
Loading…
Extractions
What the character learned after each conversation: new memories, profile updates, knowledge-graph links and mood shifts.
Select a character — or tap + New to build one.
—
Enable memory systems, then choose which sections reach
/context and in what order. Set k to 0 to skip automatic retrieval; MCP and tools still work.Maximum tokens across all recalled memory sections. Leave blank for unlimited; 0 omits memories from context.
every
user turns
Learning runs once a chat has this many new user turns since its last extraction.
POST /extract forces it early.Emotion defaults
Set the character’s resting mood and initial feelings toward new users. Existing learned mood and relationships are preserved. Save with “Save memories & order” above.
Relationship defaults · −1 to 1
Deduplication & decision model
Reconcile duplicate structured memories after extraction. Save with “Save memories & order” above. When disabled, the legacy exact-match guard still applies.
Context order
Numbered items define
context_order and context_text. Empty recalled sections are skipped per request.
token budget
Enable World under Persona & memories, save, then open this tab.
Add and edit the places, people, schedules, and facts that make up this world.
Quick add
Advanced source editor
Edit or replace the complete world as JSON or YAML.
New character
Create a character
- 1Identity
- 2Wiki files
- 3Dialogues
- 4Memories
- 5Build
Pick a name and a short description. The name becomes the character's folder and identifier.
Drop the character's lore, story or reference docs. Markdown is chunked by header. Optional — you can add files later.
Drop dialogue transcripts. Each turn should be Speaker: text, separated by blank lines. Optional.
Pick which memories the character should maintain. Defaults are sensible; change only what you need.
Build the indexes. This chunks the files, embeds them, and (if enabled) runs knowledge-graph extraction. May take a while.
Ready to build.