gemini_adk_fluent_rs/live/

config.rs

//! Model, session, and tool configuration methods for `Live`.
//!
//! # Boolean-setter rule
//!
//! Every L2 fluent builder follows one rule for on/off capabilities:
//!
//! - A capability that is **off by default** is enabled by a no-argument verb:
//!   `.transcription()`, `.session_resume()`, `.affective_dialog()`,
//!   `.proactive_audio()`, `.include_thoughts()`, `.prompt_on_enter()`.
//! - A capability that is **on by default** is disabled by `no_<x>()`:
//!   `.no_tool_advisory()`.
//! - A `bool` parameter appears only where both values are routinely passed
//!   from data — a `SessionSpec` mapping a field onto the builder — never as
//!   the way an application spells "on".
//!
//! There is no `.x(true)` / `.x(false)` pair to remember: the method name says
//! which way it flips, and the default is the absence of the call.

use std::sync::Arc;
use std::time::Duration;

use gemini_adk_rs::live::needs::RepairConfig;
use gemini_adk_rs::live::persistence::SessionPersistence;
use gemini_adk_rs::live::steering::{ContextDelivery, SteeringMode};
use gemini_adk_rs::live::{Delivery, DeliveryConfig, ResultFormatter, ToolExecutionMode};
use gemini_adk_rs::tool::{ToolDispatcher, ToolFunction};
use gemini_genai_rs::prelude::*;

use super::{DeferredAgentTool, Live};

/// Token count at which the default context window compression kicks in.
///
/// Every Live model both platforms serve today has a 128k-token window, and
/// audio runs at roughly 25 tokens a second in each direction, so this fires
/// only in a call that is already about an hour old — never in a short or
/// medium session — while leaving headroom before the hard limit.
pub const DEFAULT_COMPRESSION_TRIGGER_TOKENS: u32 = 100_000;

/// Token count the default sliding window compresses the history down to.
///
/// Half the trigger: enough to keep the last half hour or so of a call in
/// full, so the model still has the conversation it is in.
pub const DEFAULT_COMPRESSION_TARGET_TOKENS: u32 = 50_000;

impl Live {
    // -- Model & Voice --

    /// Set the Gemini model.
    ///
    /// Without this, connect resolves a default the target platform actually
    /// serves (overridable via the `GEMINI_MODEL` environment variable) —
    /// see the `connect_from_env` docs on [`Live`].
    pub fn model(mut self, model: ModelId) -> Self {
        self.config = self.config.model(model);
        self
    }

    /// Set the output voice.
    pub fn voice(mut self, voice: Voice) -> Self {
        self.config = self.config.voice(voice);
        self
    }

    /// Set the system instruction.
    pub fn instruction(mut self, instruction: impl Into<String>) -> Self {
        self.config = self.config.system_instruction(instruction);
        self
    }

    /// Switch to text-only mode (no audio output).
    ///
    /// Sets response modality to `Text` and disables speech config.
    /// Use with `ModelId::LIVE_2_5_FLASH_NATIVE_AUDIO` for text-only conversations.
    pub fn text_only(mut self) -> Self {
        self.config = self.config.text_only();
        self
    }

    /// Add a raw `Tool` declaration to the session configuration.
    ///
    /// Use this for tools that aren't registered through the `ToolDispatcher`
    /// (e.g., raw `FunctionDeclaration` lists, Google Search, code execution).
    pub fn add_tool(mut self, tool: Tool) -> Self {
        self.config = self.config.add_tool(tool);
        self
    }

    /// Set a greeting prompt to trigger the model to initiate the conversation.
    ///
    /// When set, this text is sent immediately after the session connects,
    /// causing the model to respond first (e.g. with a greeting or introduction).
    ///
    /// ```no_run
    /// # use gemini_adk_fluent_rs::prelude::*;
    /// # async fn run() -> Result<(), AgentError> {
    /// let handle = Live::builder()
    ///     .instruction("You are a friendly assistant")
    ///     .greeting("Greet the user warmly and introduce yourself.")
    ///     .connect_from_env()
    ///     .await?;
    /// // Model will speak first without any user input
    /// # let _ = handle; Ok(())
    /// # }
    /// ```
    pub fn greeting(mut self, prompt: impl Into<String>) -> Self {
        self.greeting = Some(prompt.into());
        self
    }

    /// Set the temperature.
    pub fn temperature(mut self, temp: f32) -> Self {
        self.config = self.config.temperature(temp);
        self
    }

    // -- Wire recording --

    /// Record every wire byte (both directions) to a JSONL log at `path`.
    ///
    /// The log is written by a
    /// [`FileWireRecorder`](gemini_genai_rs::transport::FileWireRecorder) created at connect time (a connect error is returned if the file cannot
    /// be created). Replay it offline with `adk session replay <path>` or
    /// [`gemini_adk_rs::live::replay::replay_session`].
    ///
    /// ```no_run
    /// # use gemini_adk_fluent_rs::prelude::*;
    /// # async fn run() -> Result<(), AgentError> {
    /// let handle = Live::builder()
    ///     .record_wire("/tmp/session.wire.jsonl")
    ///     .connect_from_env()
    ///     .await?;
    /// # let _ = handle; Ok(())
    /// # }
    /// ```
    pub fn record_wire(mut self, path: impl Into<std::path::PathBuf>) -> Self {
        self.record_wire_path = Some(path.into());
        self
    }

    /// Record every wire byte to a custom
    /// [`WireRecorder`](gemini_genai_rs::transport::WireRecorder) implementation.
    ///
    /// Overrides (and is overridden by) the most recent of this and
    /// [`record_wire`](Self::record_wire).
    pub fn wire_recorder(
        mut self,
        recorder: Arc<dyn gemini_genai_rs::transport::WireRecorder>,
    ) -> Self {
        self.record_wire_path = None;
        self.config = self.config.record_wire(recorder);
        self
    }

    // -- Tools --

    /// Register tools: a `+`-combined [`ToolComposite`] from the `T`
    /// namespace, or a single [`ToolFunction`] (a `SimpleTool`/`TypedTool`,
    /// the value a `#[tool]` function returns, an `Arc<dyn ToolFunction>`).
    ///
    /// Runtime tools go into the session's dispatcher (created on demand),
    /// built-ins into the session config, agent tools share the session
    /// `State`, and `T::mcp(..)` toolsets are connected at `connect`.
    ///
    /// ```no_run
    /// # use gemini_adk_fluent_rs::prelude::*;
    /// #[derive(serde::Deserialize, schemars::JsonSchema)]
    /// struct City {
    ///     /// The city to report on.
    ///     city: String,
    /// }
    ///
    /// Live::builder()
    ///     .tools(
    ///         T::typed("get_weather", "Get weather", |args: City| async move {
    ///             Ok(serde_json::json!({ "city": args.city, "temp": 22 }))
    ///         })
    ///         + T::google_search()
    ///     );
    /// ```
    ///
    /// [`ToolComposite`]: crate::compose::tools::ToolComposite
    pub fn tools(mut self, tools: impl Into<crate::compose::tools::ToolComposite>) -> Self {
        use crate::compose::tools::ToolResolution;
        for entry in tools.into().entries {
            match entry.classify() {
                ToolResolution::Runtime(f) => {
                    self.dispatcher
                        .get_or_insert_with(ToolDispatcher::new)
                        .register_function(f);
                }
                ToolResolution::BuiltIn(tool) => {
                    self.config = self.config.add_tool(tool);
                }
                ToolResolution::Agent {
                    name,
                    description,
                    agent,
                } => {
                    // Reuse the deferred agent-tool path so the sub-agent shares
                    // the session State created at connect time.
                    self.deferred_agent_tools.push(DeferredAgentTool {
                        name,
                        description,
                        agent,
                    });
                }
                ToolResolution::Deferred(deferred) => {
                    // MCP toolsets need an async connection; resolve them at
                    // connect time (see build_and_connect).
                    self.deferred_tools.push(deferred);
                }
            }
        }
        self
    }

    /// Register one tool — anything that implements [`ToolFunction`].
    ///
    /// ```no_run
    /// # use gemini_adk_fluent_rs::prelude::*;
    /// #[tool("Get the weather for a city")]
    /// async fn get_weather(city: String) -> Result<serde_json::Value, ToolError> {
    ///     Ok(serde_json::json!({"city": city, "temp": 22}))
    /// }
    /// Live::builder().tool(get_weather());
    /// ```
    pub fn tool(self, f: impl ToolFunction + 'static) -> Self {
        self.tools(crate::compose::tools::ToolComposite::from_function(
            Arc::new(f),
        ))
    }

    /// Use a [`ToolDispatcher`] you built yourself as the session's dispatcher
    /// — the escape hatch for streaming tools, input-streaming tools, or a
    /// dispatcher shared with other components.
    ///
    /// Tools already registered through [`tools`](Self::tools) or
    /// [`tool`](Self::tool) are kept: they are merged into `dispatcher`, whose
    /// own tools, timeout and confirmation provider win on a name clash.
    /// Tools registered after this call are added to it.
    pub fn dispatcher(mut self, mut dispatcher: ToolDispatcher) -> Self {
        if let Some(registered) = self.dispatcher.take() {
            dispatcher.merge(registered);
        }
        self.dispatcher = Some(dispatcher);
        self
    }

    /// Register a text agent as a tool the live model can call.
    ///
    /// The agent shares the session's `State`, so it can read live-extracted
    /// values and its mutations are visible to watchers and phase transitions.
    ///
    /// ```no_run
    /// # use gemini_adk_fluent_rs::prelude::*;
    /// # let verifier_agent = FnTextAgent::new("verifier", |_| Ok("verified".to_string()));
    /// # let calc_pipeline = FnTextAgent::new("calc", |_| Ok("plan".to_string()));
    /// Live::builder()
    ///     .agent_tool("verify_identity", "Verify caller identity", verifier_agent)
    ///     .agent_tool("calc_payment", "Calculate payment plans", calc_pipeline);
    /// ```
    pub fn agent_tool(
        mut self,
        name: impl Into<String>,
        description: impl Into<String>,
        agent: impl gemini_adk_rs::text::TextAgent + 'static,
    ) -> Self {
        self.deferred_agent_tools.push(DeferredAgentTool {
            name: name.into(),
            description: description.into(),
            agent: Arc::new(agent),
        });
        self
    }

    /// Register a text agent (already `Arc`'d) as a tool.
    #[deprecated(
        since = "2.1.0",
        note = "`agent_tool` accepts an `Arc<dyn TextAgent>` directly; use `agent_tool`"
    )]
    pub fn agent_tool_arc(
        mut self,
        name: impl Into<String>,
        description: impl Into<String>,
        agent: Arc<dyn gemini_adk_rs::text::TextAgent>,
    ) -> Self {
        self.deferred_agent_tools.push(DeferredAgentTool {
            name: name.into(),
            description: description.into(),
            agent,
        });
        self
    }

    /// Enable Google Search built-in tool.
    pub fn google_search(mut self) -> Self {
        self.config = self.config.with_google_search();
        self
    }

    /// Enable code execution built-in tool.
    pub fn code_execution(mut self) -> Self {
        self.config = self.config.with_code_execution();
        self
    }

    /// Enable URL context built-in tool.
    pub fn url_context(mut self) -> Self {
        self.config = self.config.with_url_context();
        self
    }

    /// Mark a tool for background execution (zero dead-air).
    ///
    /// When the model calls this tool, an immediate "running" acknowledgment
    /// is sent back while the tool executes in a background task. The final
    /// result is delivered asynchronously when complete.
    pub fn tool_background(mut self, tool_name: impl Into<String>) -> Self {
        self.tool_execution_modes.insert(
            tool_name.into(),
            ToolExecutionMode::Background {
                formatter: None,
                scheduling: None,
            },
        );
        self
    }

    /// Mark a tool for background execution with a custom result formatter.
    ///
    /// The formatter controls the shape of the acknowledgment ("running"),
    /// completion, and cancellation messages sent to the model.
    pub fn tool_background_with_formatter(
        mut self,
        tool_name: impl Into<String>,
        formatter: Arc<dyn ResultFormatter>,
    ) -> Self {
        self.tool_execution_modes.insert(
            tool_name.into(),
            ToolExecutionMode::Background {
                formatter: Some(formatter),
                scheduling: None,
            },
        );
        self
    }

    /// Mark a tool for background execution with a specific scheduling mode.
    ///
    /// The scheduling mode controls how the model handles async results:
    /// - `Interrupt`: halts current output, immediately reports the result
    /// - `WhenIdle`: waits until current output finishes before handling
    /// - `Silent`: integrates the result without notifying the user
    pub fn tool_background_with_scheduling(
        mut self,
        tool_name: impl Into<String>,
        scheduling: gemini_genai_rs::prelude::FunctionResponseScheduling,
    ) -> Self {
        self.tool_execution_modes.insert(
            tool_name.into(),
            ToolExecutionMode::Background {
                formatter: None,
                scheduling: Some(scheduling),
            },
        );
        self
    }

    // -- Audio/Video Config --

    /// Enable transcription of both the user's speech and the model's audio
    /// (delivered to `on_input_transcript` / `on_output_transcript`).
    pub fn transcription(self) -> Self {
        self.input_transcription().output_transcription()
    }

    /// Enable transcription of the user's speech only.
    pub fn input_transcription(mut self) -> Self {
        self.config = self.config.input_transcription(true);
        self
    }

    /// Enable transcription of the model's audio output only.
    pub fn output_transcription(mut self) -> Self {
        self.config = self.config.output_transcription(true);
        self
    }

    /// Enable thinking/reasoning with a token budget (Gemini 2.5+).
    ///
    /// Sets the thinking budget for the Live session. Use with
    /// `.include_thoughts()` and `.on_thought()` to receive thought summaries.
    ///
    /// ```no_run
    /// # use gemini_adk_fluent_rs::prelude::*;
    /// Live::builder()
    ///     .thinking(1024)
    ///     .include_thoughts()
    ///     .on_thought(|text| println!("[Thought] {text}"));
    /// ```
    ///
    /// **Platform support:** Google AI only. On Vertex AI, `thinkingConfig`
    /// is automatically stripped from the setup message.
    pub fn thinking(mut self, budget: u32) -> Self {
        self.config = self.config.thinking(budget);
        self
    }

    /// Set the thinking level, for models that take one instead of a budget.
    /// Gemini 3.8 Live Extended Thinking
    /// ([`ModelId::LIVE_3_8_EXTENDED_THINKING`]) refuses a session without it.
    pub fn thinking_level(mut self, level: ThinkingLevel) -> Self {
        self.config = self.config.thinking_level(level);
        self
    }

    /// Include the model's thought summaries in responses.
    ///
    /// When enabled, the model emits `SessionEvent::Thought` events containing
    /// its reasoning process. Register an `.on_thought()` callback to receive them.
    ///
    /// **Platform support:** Google AI only. Stripped on Vertex AI.
    pub fn include_thoughts(mut self) -> Self {
        self.config = self.config.include_thoughts(true);
        self
    }

    /// Enable affective dialog (emotionally expressive responses).
    pub fn affective_dialog(mut self) -> Self {
        self.config = self.config.affective_dialog(true);
        self
    }

    /// Enable proactive audio: the model may decide not to respond to input
    /// it judges not addressed to it. Pair with
    /// [`soft_turn_timeout`](Self::soft_turn_timeout).
    ///
    /// Vertex AI only: Google AI refuses a setup carrying `proactivity`, so
    /// it is left off there, as it is for Gemini 3.8 Live (always on).
    pub fn proactive_audio(mut self) -> Self {
        self.config = self.config.proactive_audio(true);
        self
    }

    /// Set media resolution for video/image input.
    pub fn media_resolution(mut self, res: MediaResolution) -> Self {
        self.config = self.config.media_resolution(res);
        self
    }

    /// Answer with Gemini 3.8 Live Avatar video (Vertex AI): synchronized
    /// 24 FPS video of a prebuilt or custom avatar, delivered to
    /// [`on_media`](Self::on_media). Sets the response modality to `VIDEO`,
    /// as the API requires. Google AI's `gemini-3.8-live` refuses `VIDEO`.
    ///
    /// ```no_run
    /// # use gemini_adk_fluent_rs::prelude::*;
    /// Live::builder()
    ///     .model(ModelId::LIVE_3_8)
    ///     .avatar(AvatarConfig::prebuilt("Ben"))
    ///     .on_media(|chunk| println!("{} bytes of {}", chunk.data.len(), chunk.mime_type));
    /// ```
    pub fn avatar(mut self, avatar: AvatarConfig) -> Self {
        self.config = self.config.avatar(avatar);
        self
    }

    /// Speak in a voice replicated from a recorded sample (Gemini 3.8 Live
    /// on Vertex AI; allow-listed customers).
    pub fn replicated_voice(mut self, voice: ReplicatedVoiceConfig) -> Self {
        self.config = self.config.replicated_voice(voice);
        self
    }

    /// Bias transcription of the user's speech toward domain terms —
    /// product names, SKUs, proper nouns (Gemini 3.8 Live). Enables input
    /// transcription.
    pub fn custom_vocabulary<S: Into<String>>(
        mut self,
        terms: impl IntoIterator<Item = S>,
    ) -> Self {
        self.config = self.config.custom_vocabulary(terms);
        self
    }

    /// Transcribe the user's speech with these settings (language hints,
    /// custom vocabulary).
    pub fn input_transcription_config(mut self, config: AudioTranscriptionConfig) -> Self {
        self.config = self.config.input_transcription_config(config);
        self
    }

    /// Transcribe the model's speech with these settings.
    pub fn output_transcription_config(mut self, config: AudioTranscriptionConfig) -> Self {
        self.config = self.config.output_transcription_config(config);
        self
    }

    // -- VAD & Activity --

    /// Run the RNNoise speech enhancer over outgoing mic audio inside
    /// `send_audio` *(feature `denoise`)* — the same stage as
    /// [`voice::Denoiser`](crate::voice::Denoiser), applied server-side to
    /// hosted surfaces (web bridge, API server) that do not run a local
    /// pump. See the hardening chapter for the measured benchmark.
    #[cfg(feature = "denoise")]
    pub fn mic_denoise(mut self) -> Self {
        self.input_audio.stages.push(InputStage::Denoise);
        self
    }

    /// Chain a [`NoiseGate`](crate::voice::NoiseGate) over outgoing mic
    /// audio inside `send_audio`, after any denoiser — frames whose RMS
    /// falls below `threshold_rms` are silenced, with `hold_frames` of
    /// hangover. Calibrate the threshold between the caller's level and the
    /// background (measured sweet spot 400–700 behind the denoiser).
    pub fn mic_noise_gate(mut self, threshold_rms: f64, hold_frames: u32) -> Self {
        self.input_audio.stages.push(InputStage::NoiseGate {
            threshold_rms,
            hold_frames,
        });
        self
    }

    /// Chain any [`InputAudioProcessor`](gemini_adk_rs::live::InputAudioProcessor)
    /// over outgoing mic audio inside `send_audio` — the open slot for
    /// application-side stages (a DeepFilterNet enhancer, AGC, a custom
    /// filter). Stages run in the order configured.
    pub fn mic_processor(
        mut self,
        processor: impl gemini_adk_rs::live::InputAudioProcessor + 'static,
    ) -> Self {
        self.input_audio
            .stages
            .push(InputStage::Custom(Box::new(processor)));
        self
    }

    /// Replace the input VAD's configuration (the detector that runs inside
    /// `send_audio` for client-side speech edges). Use
    /// [`VadConfig::noisy_street()`](gemini_genai_rs::vad::VadConfig::noisy_street)
    /// behind [`mic_denoise`](Self::mic_denoise) for noisy environments.
    pub fn input_vad(mut self, config: gemini_genai_rs::vad::VadConfig) -> Self {
        self.input_audio.vad = Some(config);
        self
    }

    /// Give this client's input VAD interruption authority: the session is
    /// configured with the server's automatic activity detection disabled,
    /// and `send_audio` emits `activityStart`/`activityEnd` on the input
    /// VAD's speech edges. Measured ~2× faster barge-in than server
    /// authority; pair with [`mic_denoise`](Self::mic_denoise) (and
    /// [`mic_noise_gate`](Self::mic_noise_gate)) or noise will drive the
    /// marks.
    pub fn client_interruption_authority(mut self) -> Self {
        self.input_audio.client_authority = true;
        self
    }

    /// Install a turn-commit policy between the input VAD's speech edges and
    /// the activity marks sent under
    /// [`client_interruption_authority`](Self::client_interruption_authority).
    ///
    /// Raw edges make two measured mistakes as turn signals (TurnBench dev
    /// set, 38 real dyadic conversations): committing end-of-turn during
    /// mid-turn pauses (fp 0.206 raw -> 0.087 with an 800 ms hold), and
    /// treating backchannels ("mm-hm") over model speech as barge-ins
    /// (fp 0.702 raw -> 0.062 with a 1400 ms sustain). See
    /// [`TurnCommitConfig`](gemini_adk_rs::live::TurnCommitConfig) for the
    /// presets carrying those operating points.
    pub fn turn_commit(mut self, config: gemini_adk_rs::live::TurnCommitConfig) -> Self {
        self.input_audio.turn_commit = Some(config);
        self
    }

    /// [`turn_commit`](Self::turn_commit) with millisecond knobs.
    pub fn turn_commit_ms(self, eot_hold_ms: u64, min_interruption_ms: u64) -> Self {
        self.turn_commit(gemini_adk_rs::live::TurnCommitConfig {
            eot_hold: std::time::Duration::from_millis(eot_hold_ms),
            min_interruption: std::time::Duration::from_millis(min_interruption_ms),
        })
    }

    /// Set only the end-of-turn hold, keeping (or defaulting) the sustain —
    /// one of the two granular forms Flow Studio codegen emits.
    pub fn turn_commit_eot_hold_ms(mut self, eot_hold_ms: u64) -> Self {
        let mut config = self.input_audio.turn_commit.unwrap_or_default();
        config.eot_hold = std::time::Duration::from_millis(eot_hold_ms);
        self.input_audio.turn_commit = Some(config);
        self
    }

    /// Set only the interruption sustain, keeping (or defaulting) the hold.
    pub fn turn_commit_min_interruption_ms(mut self, min_interruption_ms: u64) -> Self {
        let mut config = self.input_audio.turn_commit.unwrap_or_default();
        config.min_interruption = std::time::Duration::from_millis(min_interruption_ms);
        self.input_audio.turn_commit = Some(config);
        self
    }

    /// Configure server-side VAD.
    pub fn vad(mut self, detection: AutomaticActivityDetection) -> Self {
        self.config = self.config.server_vad(detection);
        self
    }

    /// Set activity handling mode (interrupts vs no-interruption).
    pub fn activity_handling(mut self, handling: ActivityHandling) -> Self {
        self.config = self.config.activity_handling(handling);
        self
    }

    /// Set turn coverage mode.
    pub fn turn_coverage(mut self, coverage: TurnCoverage) -> Self {
        self.config = self.config.turn_coverage(coverage);
        self
    }

    // -- Session Lifecycle --

    /// Enable session resumption: the server issues resumption handles that
    /// [`session_resume_from`](Self::session_resume_from) accepts on the next
    /// connect.
    pub fn session_resume(mut self) -> Self {
        self.config = self.config.session_resumption();
        self
    }

    /// Enable session resumption in transparent mode (Vertex AI only; left off
    /// the wire on Google AI): resumption updates
    /// also name the last client message the server consumed, so a resumed
    /// session knows what to send again.
    pub fn transparent_resumption(mut self) -> Self {
        self.config = self.config.transparent_resumption();
        self
    }

    /// Accept conversation history sent as client content before the first
    /// turn — required by Gemini 3.8 Live to seed history.
    pub fn history_in_client_content(mut self) -> Self {
        self.config = self.config.initial_history_in_client_content(true);
        self
    }

    /// Ask the server for explicit voice-activity events at the edges of
    /// user speech (Vertex AI; left off the wire on Google AI).
    pub fn explicit_vad_signal(mut self) -> Self {
        self.config = self.config.explicit_vad_signal(true);
        self
    }

    /// Resume a previous session from a server-issued resumption handle.
    ///
    /// Capture the handle from the old session before it ends — via
    /// [`LiveHandle::resume_handle`](gemini_adk_rs::live::LiveHandle::resume_handle)
    /// (e.g. inside the `on_go_away` callback) or from a persisted
    /// [`SessionSnapshot`](gemini_adk_rs::live::SessionSnapshot) — and pass it
    /// here on the next connect. Resumption stays enabled for the new session,
    /// so fresh handles keep arriving. No automatic reconnect is performed.
    pub fn session_resume_from(mut self, handle: impl Into<String>) -> Self {
        self.config = self.config.resume_from(handle);
        self
    }

    /// Set the context window compression thresholds.
    ///
    /// Compression is **on by default** at
    /// [`DEFAULT_COMPRESSION_TRIGGER_TOKENS`] / [`DEFAULT_COMPRESSION_TARGET_TOKENS`],
    /// so a long call keeps going instead of ending when the model's context
    /// fills. Use this to tune the thresholds;
    /// [`no_context_compression`](Self::no_context_compression) turns it off.
    pub fn context_compression(mut self, trigger_tokens: u32, target_tokens: u32) -> Self {
        self.config = self
            .config
            .context_window_compression(target_tokens)
            .context_window_trigger_tokens(trigger_tokens);
        self
    }

    /// Disable context window compression.
    ///
    /// Without it the server ends the session once the model's context window
    /// is full. Turn it off only when a session is known to be short, or when
    /// the whole history must stay verbatim for its full length.
    pub fn no_context_compression(mut self) -> Self {
        self.config.context_window_compression = None;
        self
    }

    // -- Control Plane --

    /// Enable soft turn detection for proactive silence awareness.
    ///
    /// When `proactiveAudio` is enabled, the model may choose not to respond.
    /// After VAD end, if the model stays silent for `timeout`, a lightweight
    /// "soft turn" updates state and fires watchers without forcing a response.
    pub fn soft_turn_timeout(mut self, timeout: Duration) -> Self {
        self.soft_turn_timeout = Some(timeout);
        self
    }

    /// Set the steering mode for how the phase machine delivers instructions.
    ///
    /// - `InstructionUpdate` (default): Replace system instruction on transition.
    /// - `ContextInjection`: Inject steering via `send_client_content`.
    /// - `Hybrid`: Instruction on transition, context injection per turn.
    pub fn steering_mode(mut self, mode: SteeringMode) -> Self {
        self.steering_mode = mode;
        self
    }

    /// Set when model-role context turns are delivered to the wire.
    ///
    /// - `Immediate` (default): Send as a single batched frame during
    ///   TurnComplete processing.
    /// - `Deferred`: Queue context and flush before the next user send
    ///   (`send_audio`/`send_text`/`send_video`).  Eliminates isolated
    ///   WebSocket frames during silence that can confuse the model.
    ///
    /// ```no_run
    /// # use gemini_adk_fluent_rs::prelude::*;
    /// Live::builder()
    ///     .steering_mode(SteeringMode::ContextInjection)
    ///     .context_delivery(ContextDelivery::Deferred)
    ///     .phase("greeting")
    ///         .instruction("Welcome the guest")
    ///         .done()
    ///     .initial_phase("greeting");
    /// ```
    pub fn context_delivery(mut self, mode: ContextDelivery) -> Self {
        self.context_delivery = mode;
        self
    }

    /// Set the fast-lane delivery (backpressure) policy for every event class.
    ///
    /// The event router forwards fast-lane frames (audio, text, transcripts,
    /// thoughts, VAD, phase) to the fast-lane consumer over a bounded channel.
    /// By default every class is [`Delivery::Lossless`] — the router awaits
    /// (`send().await`) when the channel is full, which preserves the historical
    /// behavior. Opt classes into [`Delivery::LossyDropNewest`] to drop the
    /// newest frame on overflow instead of stalling the router (and thereby
    /// stalling control-lane routing too).
    ///
    /// ```no_run
    /// # use gemini_adk_fluent_rs::prelude::*;
    /// use gemini_adk_fluent_rs::live::{Delivery, DeliveryConfig};
    /// Live::builder()
    ///     .delivery(DeliveryConfig::default()
    ///         .audio(Delivery::LossyDropNewest)
    ///         .transcript(Delivery::LossyDropNewest));
    /// ```
    pub fn delivery(mut self, delivery: DeliveryConfig) -> Self {
        self.delivery = delivery;
        self
    }

    /// Convenience: set the audio class to [`Delivery::LossyDropNewest`] so the
    /// router never blocks on a slow audio consumer, dropping the newest PCM
    /// frame on overflow. Other classes keep their current policy.
    pub fn lossy_audio(mut self) -> Self {
        self.delivery.audio = Delivery::LossyDropNewest;
        self
    }

    /// Convenience: set the transcript class to [`Delivery::LossyDropNewest`].
    /// Other classes keep their current policy.
    pub fn lossy_transcript(mut self) -> Self {
        self.delivery.transcript = Delivery::LossyDropNewest;
        self
    }

    /// Install transcript redaction: sensitive strings (card numbers,
    /// one-time codes, custom patterns) are removed at the event router,
    /// before callbacks, the transcript buffer, extraction, or persistence
    /// see the text.
    ///
    /// ```no_run
    /// # use gemini_adk_fluent_rs::prelude::*;
    /// use gemini_adk_rs::live::redaction::TranscriptRedactor;
    /// Live::builder()
    ///     .redaction(TranscriptRedactor::new().card_numbers().long_digits(6));
    /// ```
    pub fn redaction(
        mut self,
        redactor: gemini_adk_rs::live::redaction::TranscriptRedactor,
    ) -> Self {
        self.redactor = Some(redactor);
        self
    }

    /// Enable the conversation repair protocol.
    ///
    /// Tracks unfulfilled `needs` per phase. After `nudge_after` stalled turns,
    /// injects a gentle nudge. After `escalate_after` turns, sets
    /// `repair:escalation` in state for phase guards to handle.
    pub fn repair(mut self, config: RepairConfig) -> Self {
        self.repair_config = Some(config);
        self
    }

    /// Set a session persistence backend for surviving process restarts.
    pub fn persistence(mut self, backend: Arc<dyn SessionPersistence>) -> Self {
        self.persistence = Some(backend);
        self
    }

    /// Set the session ID for persistence.
    pub fn session_id(mut self, id: impl Into<String>) -> Self {
        self.session_id = Some(id.into());
        self
    }

    /// Disable the tool availability advisory on phase transitions.
    ///
    /// By default the SDK injects a model-role context turn telling the model
    /// which tools are available in the new phase.
    pub fn no_tool_advisory(mut self) -> Self {
        self.tool_advisory = false;
        self
    }
}

/// Input-audio hardening configuration accumulated by the builder and
/// applied to the [`LiveHandle`](gemini_adk_rs::live::LiveHandle) right
/// after connect. Stages run over each outgoing frame **in the order they
/// were configured** — chain the denoiser before the gate so the gate
/// calibrates on clean levels (see `Live::mic_denoise`,
/// `Live::mic_noise_gate`, `Live::mic_processor`, `Live::input_vad`,
/// `Live::client_interruption_authority`).
#[derive(Default)]
pub struct InputAudioConfig {
    /// Ordered mic-chain stages.
    pub stages: Vec<InputStage>,
    /// Replacement input-VAD configuration.
    pub vad: Option<gemini_genai_rs::vad::VadConfig>,
    /// Client VAD sends activity marks; server auto-detection is disabled.
    pub client_authority: bool,
    /// Turn-commit policy between VAD edges and activity marks.
    pub turn_commit: Option<gemini_adk_rs::live::TurnCommitConfig>,
}

/// One stage of the input mic chain — the named stages the SDK ships plus
/// an open [`Custom`](Self::Custom) slot for any
/// [`InputAudioProcessor`](gemini_adk_rs::live::InputAudioProcessor).
pub enum InputStage {
    /// RNNoise speech enhancement (feature `denoise`).
    #[cfg(feature = "denoise")]
    Denoise,
    /// Level gate: silences frames whose RMS falls below the threshold.
    NoiseGate {
        /// RMS threshold in sample units.
        threshold_rms: f64,
        /// Quiet frames the gate stays open after the last loud one.
        hold_frames: u32,
    },
    /// Any caller-supplied processor (denoisers, AGC, custom filters).
    Custom(Box<dyn gemini_adk_rs::live::InputAudioProcessor>),
}

/// Everything [`InputAudioConfig::build_processors`] hands the handle:
/// materialized mic-chain stages, optional VAD replacement, client
/// activity authority, and the optional turn-commit policy.
pub(crate) type BuiltInputAudio = (
    Vec<Box<dyn gemini_adk_rs::live::InputAudioProcessor>>,
    Option<gemini_genai_rs::vad::VadConfig>,
    bool,
    Option<gemini_adk_rs::live::TurnCommitConfig>,
);

impl InputAudioConfig {
    /// Whether any part of the input path is configured.
    pub fn is_configured(&self) -> bool {
        !self.stages.is_empty()
            || self.vad.is_some()
            || self.client_authority
            || self.turn_commit.is_some()
    }

    /// Materialize the configured stages into runnable processors.
    pub(crate) fn build_processors(self) -> BuiltInputAudio {
        let processors = self
            .stages
            .into_iter()
            .map(
                |stage| -> Box<dyn gemini_adk_rs::live::InputAudioProcessor> {
                    match stage {
                        #[cfg(feature = "denoise")]
                        InputStage::Denoise => Box::new(crate::voice::Denoiser::new(16_000)),
                        InputStage::NoiseGate {
                            threshold_rms,
                            hold_frames,
                        } => Box::new(crate::voice::NoiseGate::new(threshold_rms, hold_frames)),
                        InputStage::Custom(processor) => processor,
                    }
                },
            )
            .collect();
        (
            processors,
            self.vad,
            self.client_authority,
            self.turn_commit,
        )
    }
}

#[cfg(test)]
mod input_audio_tests {
    use super::*;

    #[test]
    fn turn_commit_flows_through_build() {
        let config = InputAudioConfig {
            turn_commit: Some(gemini_adk_rs::live::TurnCommitConfig::conversational()),
            ..Default::default()
        };
        assert!(config.is_configured());
        let (_, _, _, turn_commit) = config.build_processors();
        assert_eq!(
            turn_commit,
            Some(gemini_adk_rs::live::TurnCommitConfig::conversational())
        );
    }

    struct Doubler;
    impl gemini_adk_rs::live::InputAudioProcessor for Doubler {
        fn process_frame(&mut self, frame: &mut Vec<i16>) {
            for s in frame.iter_mut() {
                *s = s.saturating_mul(2);
            }
        }
    }

    #[test]
    fn stages_run_in_configured_order() {
        let live = crate::live::Live::builder()
            .mic_processor(Doubler)
            .mic_noise_gate(400.0, 3)
            .input_vad(gemini_genai_rs::vad::VadConfig::noisy_street())
            .client_interruption_authority();
        assert!(live.input_audio.is_configured());
        assert_eq!(live.input_audio.stages.len(), 2);
        assert!(matches!(live.input_audio.stages[0], InputStage::Custom(_)));
        assert!(matches!(
            live.input_audio.stages[1],
            InputStage::NoiseGate { .. }
        ));
        let (mut processors, vad, client, _) = live.input_audio.build_processors();
        assert_eq!(processors.len(), 2);
        assert_eq!(vad.unwrap().start_threshold_db, 21.0);
        assert!(client);
        // The custom stage actually runs: 100 doubles to 200, then the gate
        // (RMS 200 < 400) silences the frame — order is observable.
        let mut frame = vec![100i16; 480];
        for p in processors.iter_mut() {
            p.process_frame(&mut frame);
        }
        assert!(
            frame.iter().all(|&s| s == 0),
            "gate should silence the doubled but still-quiet frame"
        );
    }
}

#[cfg(test)]
mod compression_default_tests {
    use super::*;
    use gemini_genai_rs::prelude::{ContextWindowCompressionConfig, SlidingWindow};

    #[test]
    fn compression_is_on_by_default_at_the_documented_thresholds() {
        let live = Live::builder();
        assert_eq!(
            live.config.context_window_compression,
            Some(ContextWindowCompressionConfig {
                sliding_window: Some(SlidingWindow {
                    target_tokens: Some(DEFAULT_COMPRESSION_TARGET_TOKENS),
                }),
                trigger_tokens: Some(DEFAULT_COMPRESSION_TRIGGER_TOKENS),
            })
        );
    }

    #[test]
    fn explicit_thresholds_replace_the_default() {
        let live = Live::builder().context_compression(4096, 2048);
        let cwc = live
            .config
            .context_window_compression
            .expect("compression set");
        assert_eq!(cwc.trigger_tokens, Some(4096));
        assert_eq!(cwc.sliding_window.and_then(|w| w.target_tokens), Some(2048));
    }

    #[test]
    fn no_context_compression_turns_it_off() {
        let live = Live::builder().no_context_compression();
        assert_eq!(live.config.context_window_compression, None);
    }
}