All docs
Docs
Core Concepts

Storage profiles

Choose row, log, searchable, and archive-oriented table behavior for agent-native workloads.

Overview

Storage profiles are table-level physical-behavior hints. They tell Synapsor how a table is expected to behave for hot lookups, stable reference data, append-heavy history, audit evidence, searchable knowledge, or colder archive/reporting data.

Physical layout in the current single-node engine is explicit: hot_state, reference_data, append_log, and audit_log are tuple/row-oriented profiles; searchable_knowledge is the hybrid profile with tuple rows plus lexical/vector-style side indexes; columnar_archive is the column-oriented archive profile.

Profiles are not authorization rules and they do not make a table part of an evidence bundle by themselves. Evidence is selected by agent contexts, capability plans, hybrid searches, and explicit evidence mappings.

Hosted V2 remains single-node beta software. Storage profiles help developers express intent inside the current engine; they are not serverless storage classes, HA promises, or compliance guarantees.

Profile chooser

Pick the profile that matches how the table changes and how agents will use it.

Current state
  • hot_state for active workflow rows
  • reference_data for stable dimensions
  • Best for point lookups and frequent updates
History
  • append_log for event streams
  • audit_log for evidence, approvals, replay, and decisions
  • Best for traceability and retention
Search and archive
  • searchable_knowledge for policy/docs/memory chunks
  • columnar_archive for closed history and reporting
  • Best for retrieval or cold scans

Physical layout map

Use tuple profiles for live mutable state and history, the hybrid profile for searchable text/memory, and columnar_archive for cold reporting data.

Tuple / row profiles
  • hot_state: active mutable workflow rows, sessions, current tickets, current expenses
  • reference_data: stable dimensions such as tenants, employees, products, policies by id
  • append_log: event streams, turn logs, card transactions, usage events
  • audit_log: evidence, approvals, replay records, proposal events, settlement history
Hybrid profile
  • searchable_knowledge: tuple rows plus hybrid retrieval indexes
  • Use lexical_index for text matching and vector_index for embedding/vector-style retrieval
  • Use filter_keys and zone_map for tenant, topic, role, status, region, or effective-date filters
  • Best for policy chunks, governed memory facts, docs, and guardrail text
Column profile
  • columnar_archive: column-oriented closed history and reporting snapshots
  • COMPACT TABLE ... INTO COLUMNAR moves eligible older history into a denser layout
  • ARCHIVE / EXPORT / RESTORE ARCHIVE handle colder lifecycle operations
  • Do not use columnar_archive for frequently updated hot workflow state

Developer notes

  • Use hot_state for rows that represent what is true right now.
  • Use audit_log for evidence, approvals, proposal events, and replayable decision history.
  • Use searchable_knowledge for policy chunks, memory facts, docs, and guardrail text used by hybrid search.
  • Use columnar_archive for closed history that is kept mostly for reporting or retention.
  • Remember the physical mapping: tuple/row profiles for hot_state/reference_data/append_log/audit_log, hybrid profile for searchable_knowledge, and column profile for columnar_archive.