config

Documentation

Overview

Package config loads neo's single configuration file (neo.yaml / ~/.neo/config.yaml / embedded default).

Index

• Constants
• type Backend
• type Compaction
• type Config
  • func Load() (*Config, error)
  • func (c *Config) AgentsFileEnabled() bool
  • func (c *Config) NamedPhases() []phase.Definition
  • func (c *Config) PromptCachingEnabled() bool
  • func (c *Config) SkillsEnabled() bool
  • func (c *Config) Source() string
  • func (c *Config) SubagentsConfigured() bool
  • func (c *Config) SubscriptionAuth() bool
  • func (c *Config) VerboseEnabled() bool
• type Features
• type Output

Constants

const (

	// OpenAI auth modes (the openai_auth config key).
	OpenAIAuthAPIKey       = "api_key"
	OpenAIAuthSubscription = "subscription"
)

Resolution paths and embedded fallback.

Types

type Backend

type Backend struct {
	Provider string `yaml:"provider"`
	Model    string `yaml:"model"`
}

Backend optionally pins chat-spawned subagents to a provider and model. A zero value means subagents follow the coordinator's active backend.

type Compaction

type Compaction struct {
	// ContextWindowTokens is an optional manual override for the active model's
	// context window. When omitted, chat startup uses the compact package's
	// conservative default.
	ContextWindowTokens int `yaml:"context_window_tokens"`
}

Compaction configures when long transcripts are summarized.

type Config

type Config struct {
	// Provider selects the LLM backend: "anthropic" (default), "openai", "openrouter", or "google".
	Provider string `yaml:"provider"`
	// OpenAIAuth selects how the "openai" provider authenticates: "api_key"
	// (default, uses OPENAI_API_KEY) or "subscription" (ChatGPT/Codex
	// device-code credentials via `neo login`). This also applies when OpenAI
	// is selected through /model while another provider is the startup default.
	OpenAIAuth string     `yaml:"openai_auth"`
	Model      string     `yaml:"model"`
	Subagents  Backend    `yaml:"subagents"`
	Features   Features   `yaml:"features"`
	Compaction Compaction `yaml:"compaction"`
	// ToolApprovals lists optional interactive tool names and Bash prefixes.
	ToolApprovals []string `yaml:"tool_approvals"`
	Output        Output   `yaml:"output"`
	// Phases overrides Neo's built-in named prompts by name or adds custom
	// prompts. Resolution always retains defaults that are not overridden.
	Phases map[string]phase.Definition `yaml:"phases"`
	// contains filtered or unexported fields
}

Config is the parsed neo.yaml.

func Load

func Load() (*Config, error)

Load reads the first available config file:

./neo.yaml → ~/.neo/config.yaml → embedded default

First hit wins — no merging.

func (*Config) AgentsFileEnabled

func (c *Config) AgentsFileEnabled() bool

AgentsFileEnabled reports whether AGENTS.md loading is on (default: true).

func (*Config) NamedPhases

func (c *Config) NamedPhases() []phase.Definition

NamedPhases returns Neo's defaults with configured definitions overlaid. Config loading validates this data, so the error is unreachable for loaded configurations. Programmatically constructed zero values still get defaults.

func (*Config) PromptCachingEnabled

func (c *Config) PromptCachingEnabled() bool

PromptCachingEnabled reports whether the static system prompt is marked for provider-side prompt caching (default: true).

func (*Config) SkillsEnabled

func (c *Config) SkillsEnabled() bool

SkillsEnabled reports whether skill loading is on (default: true).

func (*Config) Source

func (c *Config) Source() string

Source describes where this Config was loaded from (a file path or "embedded"). Useful in error messages and diagnostics.

func (*Config) SubagentsConfigured

func (c *Config) SubagentsConfigured() bool

SubagentsConfigured reports whether subagents use a backend independent of the coordinator. parseConfig fills either missing half when one is supplied.

func (*Config) SubscriptionAuth

func (c *Config) SubscriptionAuth() bool

SubscriptionAuth reports whether the openai provider should authenticate via a ChatGPT/Codex subscription rather than an API key.

func (*Config) VerboseEnabled

func (c *Config) VerboseEnabled() bool

VerboseEnabled reports whether full tool output rendering is on (default: false).

type Features

type Features struct {
	AgentsFile    *bool `yaml:"agents_file"`    // load AGENTS.md into the chat system prompt
	Skills        *bool `yaml:"skills"`         // discover and expand $name and /name skills
	PromptCaching *bool `yaml:"prompt_caching"` // cache the static system prompt prefix
}

Features toggles optional, layered capabilities. The core agent loop is never gated by a feature — only capabilities built on top of it. Each flag is a tri-state *bool: nil ("absent from config") falls back to a built-in default, so a minimal neo.yaml still gets the full experience; set a flag to false to turn the capability off explicitly.

type Output

type Output struct {
	// Verbose restores the full tool call/result rendering (complete file
	// contents, command output, etc). Defaults to false: Neo shows live activity
	// and concise completed receipts while still surfacing errors in full.
	Verbose *bool `yaml:"verbose"`
}

Output configures how Neo renders tool activity during a chat session.