Reasoning Configuration

Customize how Honcho reasons over peers, sessions, and messages

Honcho’s reasoning can be configured at multiple levels to control how it processes messages, generates conclusions, creates summaries, and builds peer representations.

Configuration follows a hierarchy: message > session > workspace > global defaults. Settings at lower levels override those at higher levels, giving you fine-grained control over behavior.

Configuration Hierarchy

Honcho uses a hierarchical configuration system where more specific settings override more general ones:

1. Global Defaults: Built-in system defaults
2. Workspace Configuration: Settings that apply to all sessions in a workspace
3. Session Configuration: Settings that apply to all messages in a session
4. Message Configuration: Settings that apply to a specific message

Separately, you can configure the reasoning status of a peer. This overrides defaults and workspace configuration, but not session or message configuration.

All configuration fields are optional. If not specified, the value is inherited from the next level up in the hierarchy.

Configuration Options

Reasoning Configuration

Controls whether the system should reason over messages.

Field	Type	Description
enabled	bool	Whether to enable reasoning functionality. When disabled, no facts or representations are generated.

Python

from honcho import Honcho

honcho = Honcho()

# Disable reasoning at session level
session = honcho.session("private-session", config={
    "reasoning": {"enabled": False}
})

Peer Card Configuration

Controls how peer cards (containing key biographical information) are generated and used.

Field	Type	Description
use	bool	Whether to use peer cards during the reasoning process.
create	bool	Whether to generate and update peer cards based on message content.

Python

# Disable peer card generation but still use existing cards
session = honcho.session("my-session", config={
    "peer_card": {"create": False, "use": True}
})

Summary Configuration

Controls automatic conversation summarization. Available at workspace and session levels only.

Field	Type	Description
enabled	bool	Whether to enable summary functionality.
messages_per_short_summary	int	Number of messages between short summaries. Must be ≥ 10.
messages_per_long_summary	int	Number of messages between long summaries. Must be ≥ 20 and greater than messages_per_short_summary.

Python

# Customize summary frequency
session = honcho.session("verbose-session", config={
    "summary": {
        "enabled": True,
        "messages_per_short_summary": 15,
        "messages_per_long_summary": 45
    }
})

Dream Configuration

Controls the “dreaming” process that consolidates and refines representations. Available at workspace and session levels only.

Field	Type	Description
enabled	bool	Whether to enable dream functionality. Automatically disabled if reasoning is disabled.

Python

# Disable dreams for a workspace
honcho.set_configuration({
    "dream": {
        "enabled": False
    }
})

Custom Instructions

Custom instructions are free-text guidance added to the prompts Honcho’s agents use, for example “Always write in German” or “Only track durable work preferences.” Set one shared value for every agent, override it per agent, or both.

Field	Applies to
custom_instructions (top level)	Shared fallback for the deriver, summarizer, dialectic, and dreamer
reasoning.custom_instructions	Deriver (conclusion extraction)
summary.custom_instructions	Session summaries
dialectic.custom_instructions	The peer and workspace chat endpoints
dream.custom_instructions	Dreaming (deductive and inductive conclusions)
peer_card.custom_instructions	Peer card updates during dreaming. Added on top of the dream instructions, applied within the peer card’s allowed entry kinds, and never falls back to the shared value.

Resolution rules:

1. The most specific level wins. Honcho checks session, then workspace configuration and uses the first level that sets a value, so a session’s shared instructions beat a workspace’s summary.custom_instructions. Messages can only set reasoning.custom_instructions, which overrides everything else for the deriver (see Message Configuration).
2. Within a level, a module’s own value beats the shared value.
3. An empty string means “none”. Set "" to turn off inherited instructions for a module at that level.
4. Other sections never fall back to reasoning. reasoning.custom_instructions only affects the deriver, as it always has.

Each value is limited to DERIVER_MAX_CUSTOM_INSTRUCTIONS_TOKENS (default 2000).

Python

# German everywhere, bullet-point summaries, and no instructions for the dialectic
honcho.set_configuration({
    "custom_instructions": "Always write in German.",
    "summary": {"custom_instructions": "Write summaries as bullet points in German."},
    "dialectic": {"custom_instructions": ""},
})

---

Peer Configuration

By default, all peers are “observed” by Honcho. This means that Honcho will reason over messages sent by the peer and generate a representation of them. In most cases, this is why you use Honcho! However, sometimes an application requires a peer that should not be observed: for example, an assistant or game NPC that your program will never need to access advanced reasoning for.

You may therefore disable observation of a peer by setting the observe_me flag in their configuration to false.

If the peer has a session-level configuration, it will override this configuration. If the flag is not set, or is set to true, the peer will be observed.

For session-level observation controls and local representations (where peers build separate models of each other), see Directional Representations.

Python

from honcho import Honcho

# Initialize client
honcho = Honcho()

# Create peer with configuration
peer = honcho.peer("my-peer", configuration={"observe_me": False})

# Change peer's configuration
peer.set_configuration({"observe_me": True})

# Note: creating the same peer again will also replace the configuration
peer = honcho.peer("my-peer", configuration={"observe_me": False})

Session Configuration

Sessions support the full configuration schema. You can disable reasoning entirely for a session, customize summary behavior, or adjust peer card settings.

Python

from honcho import Honcho

# Initialize client
honcho = Honcho()

# Create session with reasoning disabled
session = honcho.session("my-session", configuration={
    "reasoning": {"enabled": False}
})

# Create session with custom summary settings
session = honcho.session("detailed-session", configuration={
    "summary": {
        "messages_per_short_summary": 10,
        "messages_per_long_summary": 30
    }
})

Message Configuration

Individual messages can override session and workspace configuration for fine-grained control. This is useful for excluding specific messages from processing or adjusting behavior on a per-message basis.

Python

from honcho import Honcho

honcho = Honcho()
session = honcho.session("my-session")
user = honcho.peer("user")

# Create a message that skips the reasoning process
session.add_messages([
    user.message("This message won't be analyzed", configuration={
        "reasoning": {"enabled": False}
    })
])

# Give the deriver instructions for this message only
session.add_messages([
    user.message("Imported from our old CRM: renewal due in March", configuration={
        "reasoning": {"custom_instructions": "This is imported CRM data. Only extract account facts."}
    })
])

Full Configuration Schema Reference

Workspace & Session Configuration

{
  "custom_instructions": null,
  "reasoning": {
    "enabled": true,
    "custom_instructions": null
  },
  "peer_card": {
    "use": true,
    "create": true,
    "custom_instructions": null
  },
  "summary": {
    "enabled": true,
    "messages_per_short_summary": 20,
    "messages_per_long_summary": 60,
    "custom_instructions": null
  },
  "dream": {
    "enabled": true,
    "custom_instructions": null
  },
  "dialectic": {
    "custom_instructions": null
  }
}

Message Configuration

{
  "reasoning": {
    "enabled": true,
    "custom_instructions": null
  }
}

Message configuration only supports reasoning settings. Message-level reasoning.custom_instructions override session and workspace instructions for the deriver only. Peer card, summary, dream, dialectic, and shared custom_instructions are session/workspace-level only.