Nano, Supervisor, and Decision models
Name sessions with nano
Section titled “Name sessions with nano”Configure nano to generate titles for unnamed sessions without changing the model used for coding. Kent generates a title after the first response and any tool work it starts have completed. Automatic naming leaves existing names unchanged.
Choose an existing connection and an explicit nano model:
[nano]connection = "subscription"model = "gpt-6-luna"For a local model whose context window Kent cannot determine, also set model_context_window to its actual token limit.
Set nano.features.session_naming = false to disable automatic naming. Nano settings follow the normal configuration precedence, so a worktree can select another model or disable naming.
Review answers with the supervisor
Section titled “Review answers with the supervisor”The supervisor checks completed answers and supplies suggestions that the coding agent can apply in a follow-up. The answer is visible without waiting for review.
By default, the supervisor runs after answers that include successful file edits and uses the coding agent’s model. Set reviewer.model to choose another model, or reviewer.frequency = "all" to include answers without file edits. Set the frequency to "off" to disable it. Use /supervisor to toggle it during a session.
[reviewer]frequency = "edits"model = "gpt-6-astra"Set reviewer.verbose_output = true to show the full supervisor feedback in the terminal transcript.
Configuration reference
Section titled “Configuration reference”| Key | Type | Default | Usage |
|---|---|---|---|
nano.connection |
string or string array | required | Existing connection name, or equivalent connections to rotate between. |
nano.model |
string | required | Model used for automatic naming. |
nano.thinking_level |
string | model default, otherwise medium |
Reasoning effort supported by the selected model. |
nano.model_verbosity |
string | provider default | low, medium, or high. |
nano.model_context_window |
int | model context window | Required if Kent cannot determine the window. Minimum 2500. |
nano.timeout_seconds |
int | 120 |
Inference timeout, including retries, starting when capacity is available. Must be positive. |
nano.concurrency |
int | 5 |
Simultaneous nano calls across workspaces. Set only in global configuration. Must be positive. |
nano.features.session_naming |
bool | true |
Automatic naming when nano is configured. |
nano.model_capabilities.supports_reasoning_effort |
bool | model capability | Override reasoning-effort support for a custom model. |
nano.model_capabilities.supports_vision_inputs |
bool | model capability | Override image and PDF input support for a custom model. |
Supervisor
Section titled “Supervisor”Supervisor settings use the [reviewer] table. Omitted model controls inherit the corresponding coding-agent settings.
| Key | Type | Default | Environment variable | Usage |
|---|---|---|---|---|
reviewer.connection |
string | session’s selected connection | One existing connection name. | |
reviewer.frequency |
string | edits |
KENT_REVIEWER_FREQUENCY |
off, all, or edits. |
reviewer.model |
string | inherits model |
KENT_REVIEWER_MODEL |
Model used for review. |
reviewer.thinking_level |
string | inherits thinking_level |
KENT_REVIEWER_THINKING_LEVEL |
Reasoning effort supported by the selected model. |
reviewer.model_verbosity |
string | inherits model_verbosity |
KENT_REVIEWER_MODEL_VERBOSITY |
"", low, medium, or high. |
reviewer.model_context_window |
int | inherits model_context_window |
KENT_REVIEWER_MODEL_CONTEXT_WINDOW |
Explicit context-window override. Minimum 40000. |
reviewer.system_prompt_file |
optional string | inherits system_prompt_file |
Custom prompt path relative to the configuration file that supplies it. Empty paths are invalid. | |
reviewer.timeout_seconds |
int | 120 |
KENT_REVIEWER_TIMEOUT_SECONDS |
HTTP timeout. Must be positive. |
reviewer.verbose_output |
bool | false |
KENT_REVIEWER_VERBOSE_OUTPUT |
Show full supervisor feedback in the terminal transcript. |
reviewer.model_capabilities.supports_reasoning_effort |
bool | inherits main model capability | KENT_REVIEWER_MODEL_CAPABILITIES_SUPPORTS_REASONING_EFFORT |
Override reasoning-effort support for a custom model. |
reviewer.model_capabilities.supports_vision_inputs |
bool | inherits main model capability | KENT_REVIEWER_MODEL_CAPABILITIES_SUPPORTS_VISION_INPUTS |
Override image and PDF input support for a custom model. |