Common settings
These settings appear on every space, whatever the paradigm. A space’s paradigm cannot be changed after creation. Basics Rename the space. Flow binding Bind the space to one specific Flow and every Agent in that Flow automatically reads and writes it. Leave it unbound and you attach it per Agent in the Flow builder. Scope Which request parameters partition the contents. Combineflow_id / thread_id / user_id / custom keys. See Scoping.
Access (optional) Control access itself with a partition key (auto-filter by caller identity) and a required-permissions gate. See Scoping → Access control.
Wiki
Memory agent (curator) The agent that manages the Wiki space. It decides what to keep, grow, and retire.- Model Runs extraction and consolidation. Pick from providers with keys registered in the workspace. Leave blank for a low-cost default.
- Policy prompt Free-text instructions for what to keep and discard. Example: “Keep the customer’s stable preferences and decisions, discard chitchat, and supersede the old one when a preference changes.”
- Multimodal How images, audio, and video go into memory (
reuse/reference/re-analyze).Re-analyzecosts extra. - Background writes When on, saves and deletes happen off the response path, so there’s no added response latency. Retrieval is always synchronous. Recommended on.
Index
Index Index-only settings.- Index form Vector only or Graph
- Embedding model Pick from providers with keys registered in the workspace. Leave blank for a low-cost default
- Weights How to mix relevance, recency, and importance (default 3/1/1)
- Recency decay half-life (days) Default 14 days. For domains where old facts are as valid as new ones (glossaries, settled procedures), raise the half-life to reduce recency’s influence
- Temporal validity filter When on, reads only currently-valid facts, hiding ones that were superseded or expired. Off by default
- KG grounding After the vector search, also fetches neighboring facts connected by graph relationships. Turning it on reveals three values.
- Depth How many hops to walk from the seed (default 1, max 3)
- Decay Score decay per hop (0–1)
- Cap Maximum number of neighbors to fetch
Session
Scope is typicallyflow_id + thread_id.
Session buffer
The session-only section that decides how this space stores and keeps conversations.- Store Choose the storage backend.
redisHot, volatile storage. Fast, but gone after a restart, so it suits short conversations.postgresDurable storage that survives after the session ends.both(default) Reads fast from Redis and persists to Postgres.
- Recent window How many recent turns (or tokens) to keep verbatim, without summarizing.
- Unit
turns(number of turns) ortokens(number of tokens) - Size How many to keep
- Unit
- Summarize When on, older turns fold into a rolling summary. It triggers before the window fills, so context never overflows.
- Trigger At what fraction of the window budget summarization fires (0–1, default 0.8 = at 80% full)
- Recursive summary If the summary itself overflows, makes a summary of summaries. For ultra-long conversations
- Trigger message (optional) A notice left in the conversation when summarization runs. Example: “Earlier context has been summarized.”
Summarizer
This section appears when Summarize is on. It configures the agent that produces the rolling summary.- Model The model that performs summarization, from the workspace’s providers. Leave blank for a low-cost default.
- Summarize prompt (optional) Customize how summarization works. Example: “Keep every decision and open question, drop the chitchat.”