Struct AgentConfig 

pub struct AgentConfig {

    pub name: String,
    pub description: Option<String>,
    pub model_id: String,
    pub system_prompt: Option<String>,
    pub timeout_seconds: u64,
    pub temperature: Option<f64>,
    pub max_tokens: Option<usize>,
    pub compaction_strategy: CompactionStrategy,
    pub compaction_instruction: Option<String>,
    pub context_window: usize,
    pub workspace_dir: Option<PathBuf>,
    pub output_mode: Option<String>,
    pub session_id: Option<String>,
    pub mode: Mode,
    pub provider_options: Option<ProviderOptions>,
    pub ttsr_engine: Option<Arc<TtsrEngine>>,
    pub memory: Option<Arc<dyn MemoryBackend>>,
    pub todo: Option<Arc<dyn TodoStateProvider>>,
    pub agent_pool: Option<Arc<dyn AgentPoolProvider>>,
    pub url_resolver: Option<Arc<dyn UrlResolver>>,
    pub lsp: Option<Arc<dyn LspProvider>>,
    pub max_tool_result_bytes: Option<usize>,
    pub subagent_runner: Option<Arc<dyn SubagentRunner>>,
    pub subagent_depth: u8,
    pub snapshot_store: Option<Arc<dyn SnapshotStore>>,
}

Stable since 0.63.0

Agent runtime configuration

Fields

name: String

Agent name

description: Option<String>

Agent description

model_id: String

Model ID to use

system_prompt: Option<String>

System prompt

timeout_seconds: u64

Timeout in seconds for the entire agent run

temperature: Option<f64>

Temperature for generation (0.0 to 1.0)

max_tokens: Option<usize>

Maximum tokens to generate

compaction_strategy: CompactionStrategy

Compaction strategy for long conversations

compaction_instruction: Option<String>

Custom instruction passed to the compactor

context_window: usize

Model context window size (used for threshold-based compaction)

workspace_dir: Option<PathBuf>

Working directory for file tools. Defaults to current directory if None.

output_mode: Option<String>

Output mode for agent responses.

When set, the agent extracts structured output from the final response. See OutputMode for available modes.

session_id: Option<String>

Session identity used by tools that gate behavior on liveness (e.g. the issue tool’s start/close ownership checks). When Some, this value is threaded through to crate::tools::ToolContext::session_id. None means the tool receives session_id == None and ownership-gated operations will reject the call (defensive default).

mode: Mode

Autonomy mode — Mode::Default (interactive) or Mode::Auto (autonomous; the ask tool is short-circuited and a directive reinforces autonomous operation). Default: Mode::Default.

provider_options: Option<ProviderOptions>

Per-provider options for fine-grained control.

When set, these are passed through to oxicode_ai::StreamOptions::provider_options so the provider can read provider-specific settings (e.g. Anthropic adaptive thinking, OpenAI reasoning_effort, Google thinkingConfig).

ttsr_engine: Option<Arc<TtsrEngine>>

TTSR engine for stream rule checking. When set, streaming output is checked against registered rules and violations trigger crate::agent_loop::StreamOutcome::RuleInterrupt.

memory: Option<Arc<dyn MemoryBackend>>

Memory backend for memory_* tools.

todo: Option<Arc<dyn TodoStateProvider>>

Todo state provider for the todo tool.

agent_pool: Option<Arc<dyn AgentPoolProvider>>

Agent pool for Hub display and sub-agent matching.

url_resolver: Option<Arc<dyn UrlResolver>>

URL resolver for internal protocol schemes (issue://, pr://, etc.). Threaded through to crate::agent_loop::config::AgentLoopConfig::url_resolver. When None, URL-prefixed paths are treated as regular file paths.

lsp: Option<Arc<dyn LspProvider>>

LSP provider for the lsp tool. Threaded through to crate::agent_loop::config::AgentLoopConfig::lsp. When None, the lsp tool returns an error.

max_tool_result_bytes: Option<usize>

Maximum bytes of a tool result’s text content before truncation (#28 gap 1, surfaced as #32). Threaded through to crate::agent_loop::config::AgentLoopConfig::max_tool_result_bytes.

When set, tool results exceeding this limit are truncated and a "... [truncated: N bytes omitted]" marker is appended, preventing a single large tool output from consuming the context window.

None (default) = no limit. Opt-in.

subagent_runner: Option<Arc<dyn SubagentRunner>>

In-process sub-agent runner (#28 gap 3, surfaced as #32). When set, the subagent tool prefers an in-process isolated run over shelling out. Threaded through to crate::agent_loop::config::AgentLoopConfig::subagent_runner.

subagent_depth: u8

Current sub-agent nesting depth (#28 gap 3, surfaced as #32). Default 0 (top-level). The subagent tool increments this when forking a child config to cap recursion.

snapshot_store: Option<Arc<dyn SnapshotStore>>

Snapshot store for hashline line-anchored edit mode.

When Some, the read tool records file snapshots and emits [path#TAG] headers, and the edit tool validates edits against them. When None (default), hashline anchoring is disabled and the edit tool falls back to plain text replacement.

Implementations

impl AgentConfig

pub fn new(model_id: impl Into<String>) -> AgentConfig

Create a new config with the given model ID.

pub fn with_name(self, name: impl Into<String>) -> AgentConfig

Set the agent name.

pub fn with_system_prompt(self, prompt: impl Into<String>) -> AgentConfig

Set the system prompt.

pub fn with_timeout(self, seconds: u64) -> AgentConfig

Set the timeout in seconds for the entire agent run.

pub fn with_compaction_strategy( self, strategy: CompactionStrategy, ) -> AgentConfig

Set the compaction strategy for long conversations.

pub fn with_compaction_instruction( self, instruction: impl Into<String>, ) -> AgentConfig

Set a custom instruction passed to the compactor.

pub fn with_session_id(self, session_id: impl Into<String>) -> AgentConfig

Set the session identity threaded into crate::tools::ToolContext::session_id.

Tools that gate behavior on liveness (e.g. an issue tool’s start/close ownership checks) use this to identify the caller. Leaving it None causes those tools to see an empty caller id and reject ownership-gated operations (defensive default).

pub fn with_snapshot_store(self, store: Arc<dyn SnapshotStore>) -> AgentConfig

Set the hashline snapshot store — enables line-anchored edit mode in the read/edit tools (emits [path#TAG] headers, validates edits).

Trait Implementations

impl Clone for AgentConfig

fn clone(&self) -> AgentConfig

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for AgentConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl Default for AgentConfig

fn default() -> AgentConfig

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for AgentConfig

fn deserialize<__D>( __deserializer: __D, ) -> Result<AgentConfig, <__D as Deserializer<'de>>::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for AgentConfig

fn serialize<__S>( &self, __serializer: __S, ) -> Result<<__S as Serializer>::Ok, <__S as Serializer>::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.