honcho

Cross-session memory maintains a unified user representation across runs by configuring isolated AI peer identities that share a common workspace, injecting persistent context into prompts without full chat history replay. It applies multi-profile peer isolation to keep user modeling consistent yet separated.

Context budgeting trims injected memory representations to stay within specified token caps by enforcing session-summary and representation injection rules. You configure optional context token caps during setup to control how much memory the system injects into prompts.

Honcho works with both cloud and self-hosted setups for multi-profile peer isolation. You define the host identity model by setting user and AI peer identities, then apply observation and recall configuration to keep each profile isolated within a shared workspace.

Dialectic reasoning tuning configures how often and how deeply the system processes reasoning to build user representations. You need it when troubleshooting recall performance or optimizing how observations update the persistent memory model across active sessions.

Recall modes and observation toggles are configured by selecting hybrid, context, or tools recall strategies and defining what each AI peer observes. You apply cold or warm prompt strategies alongside these settings to control how memory representations are retrieved and injected.

Cross-session memory loses context when peer isolation separates AI identities without sharing a common user view, or when observation toggles are disabled. Configuring a unified host identity model with shared workspace settings ensures consistent user representation across all isolated profiles.