Files
galaxy/app/src/ai/runtime/rig.rs
T
2026-08-05 00:56:55 -05:00

513 lines
21 KiB
Rust

use std::sync::Arc;
use futures::channel::oneshot;
use futures::{FutureExt, StreamExt};
use galaxy_agent_core::{
turn_control, AgentError, AgentEvent, AgentRuntime, MessageContent, MessageRole, StopReason,
ToolCall, ToolCallDecision, ToolEvent, ToolPolicy, ToolResult, TurnCommand, Usage,
};
use galaxy_agent_rig::{OpenAICompatibleRuntime, OpenAICompatibleRuntimeConfig};
use uuid::Uuid;
use warp_multi_agent_api::response_event::stream_finished;
use warp_multi_agent_api::{self as api, ClientAction, ResponseEvent, ToolType};
use super::rig_request::{prepare_bedrock_rig_turn, prepare_rig_turn, PreparedRigTurn};
use super::rig_tool::action_from_tool_call;
use crate::ai::agent::api::{Event, RequestParams, ResponseStream, StreamEvent};
use crate::ai::agent::AIAgentAction;
use crate::ai::bedrock::client::{BedrockClient, BedrockClientConfig};
use crate::ai::bedrock::external_config::ExternalBedrockConfig;
use crate::ai::bedrock::response_translator::{
build_add_agent_output_message, build_append_text, build_create_task, build_stream_init,
build_user_query_message,
};
use crate::ai::openai::client::OpenAIClientConfig;
use crate::ai::openai::response_translator::{build_stream_finished, StreamUsage};
use crate::ai::provider::types::{ContentPart, ConversationMessage};
use crate::server::server_api::AIApiError;
pub(crate) fn rig_openai_response_stream(
config: OpenAIClientConfig,
params: RequestParams,
supported_tools: Vec<ToolType>,
supported_cli_agent_tools: Vec<ToolType>,
cancellation_rx: oneshot::Receiver<()>,
) -> ResponseStream {
let skill_path_origin = params.session_context.skill_path_origin();
let prepared = prepare_rig_turn(&config, params, supported_tools, supported_cli_agent_tools);
let model_id = prepared.request.model.as_str().to_string();
let runtime = OpenAICompatibleRuntime::new(OpenAICompatibleRuntimeConfig {
base_url: config.base_url,
api_key: config.api_key,
model: model_id.clone(),
max_output_tokens: config.max_output_tokens.map(u64::from),
supports_system_messages: config.supports_system_messages,
});
rig_response_stream(
runtime,
prepared,
skill_path_origin,
config.max_input_tokens,
"rig_openai_compatible",
cancellation_rx,
)
}
pub(crate) async fn rig_bedrock_response_stream(
config: BedrockClientConfig,
params: RequestParams,
supported_tools: Vec<ToolType>,
supported_cli_agent_tools: Vec<ToolType>,
cancellation_rx: oneshot::Receiver<()>,
) -> anyhow::Result<ResponseStream> {
let skill_path_origin = params.session_context.skill_path_origin();
let max_context_tokens = params.context_window_limit;
let model = params.model.as_str().to_string();
let max_output_tokens = Some(64_000);
let cross_region_inference = config.cross_region_inference;
let external_config = ExternalBedrockConfig::load();
let prompt_caching = !external_config.disable_prompt_caching;
let client = BedrockClient::from_config(config).await?;
let runtime = client.rig_runtime(
model.clone(),
cross_region_inference,
prompt_caching,
max_output_tokens,
)?;
let prepared = prepare_bedrock_rig_turn(
model,
max_output_tokens,
params,
supported_tools,
supported_cli_agent_tools,
);
Ok(rig_response_stream(
runtime,
prepared,
skill_path_origin,
max_context_tokens,
"rig_bedrock",
cancellation_rx,
))
}
fn rig_response_stream<R>(
runtime: R,
prepared: PreparedRigTurn,
skill_path_origin: ai::skills::SkillPathOrigin,
max_context_tokens: Option<u32>,
stream_type: &'static str,
cancellation_rx: oneshot::Receiver<()>,
) -> ResponseStream
where
R: AgentRuntime + Send + Sync + 'static,
{
let PreparedRigTurn {
task_id,
needs_create_task,
user_query,
request: turn_request,
persistent_messages,
tool_result_archive,
messages_sent,
} = prepared;
store_messages_sent(&messages_sent, &persistent_messages);
let conversation_id = turn_request.conversation_id.clone();
let model_id = turn_request.model.as_str().to_string();
let tool_policy = ToolPolicy::new(&turn_request.tools);
let stream = async_stream::stream! {
let (control_sender, control) = turn_control();
let start_future = runtime.start_turn(turn_request, control).fuse();
let cancel_future = cancellation_rx.fuse();
futures::pin_mut!(start_future, cancel_future);
let mut agent_events = futures::select_biased! {
_ = cancel_future => {
let _ = control_sender.try_send(TurnCommand::Cancel);
match start_future.await {
Ok(stream) => stream,
Err(error) => {
yield Err(agent_error(error, stream_type));
return;
}
}
}
result = start_future => match result {
Ok(stream) => stream,
Err(error) => {
yield Err(agent_error(error, stream_type));
return;
}
},
};
let request_id = Uuid::new_v4().to_string();
let conversation_id = conversation_id.unwrap_or_else(|| Uuid::new_v4().to_string());
let mut initialized = false;
let mut current_text_message_id: Option<String> = None;
let mut current_reasoning_message_id: Option<String> = None;
let mut full_text = String::new();
let mut full_reasoning = String::new();
let mut reasoning_signature = None;
let mut proposed_tools = Vec::new();
let mut assistant_history_index = None;
let mut usage = Usage::default();
loop {
let next_event = agent_events.next().fuse();
futures::pin_mut!(next_event);
futures::select_biased! {
_ = cancel_future => {
let _ = control_sender.try_send(TurnCommand::Cancel);
}
event = next_event => {
let Some(event) = event else {
yield Err(Arc::new(AIApiError::UnexpectedEof));
return;
};
let event = match event {
Ok(event) => event,
Err(error) => {
yield Err(agent_error(error, stream_type));
return;
}
};
match event {
AgentEvent::TurnStarted { .. } => {
initialized = true;
yield Ok(StreamEvent::Response(build_stream_init(&request_id, &conversation_id)));
if needs_create_task {
yield Ok(StreamEvent::Response(build_create_task(&task_id)));
}
if let Some(user_query) = &user_query {
yield Ok(StreamEvent::Response(build_user_query_message(&task_id, user_query)));
}
}
AgentEvent::TextDelta { text } => {
full_text.push_str(&text);
if let Some(message_id) = &current_text_message_id {
yield Ok(StreamEvent::Response(build_append_text(&task_id, message_id, &text)));
} else {
let message_id = Uuid::new_v4().to_string();
yield Ok(StreamEvent::Response(build_add_agent_output_message(&task_id, &message_id, &text)));
current_text_message_id = Some(message_id);
}
}
AgentEvent::ReasoningDelta { text } => {
full_reasoning.push_str(&text);
if let Some(message_id) = &current_reasoning_message_id {
yield Ok(StreamEvent::Response(build_append_reasoning(&task_id, message_id, &text)));
} else {
let message_id = Uuid::new_v4().to_string();
yield Ok(StreamEvent::Response(build_add_reasoning(&task_id, &message_id, &text)));
current_reasoning_message_id = Some(message_id);
}
}
AgentEvent::ReasoningCompleted { text, signature } => {
if current_reasoning_message_id.is_none() && !text.is_empty() {
let message_id = Uuid::new_v4().to_string();
yield Ok(StreamEvent::Response(build_add_reasoning(&task_id, &message_id, &text)));
current_reasoning_message_id = Some(message_id);
}
if !text.is_empty() {
full_reasoning = text;
}
reasoning_signature = signature;
}
AgentEvent::UsageUpdated { usage: updated } => usage = updated,
AgentEvent::Tool {
event: ToolEvent::Proposed { call },
} => {
proposed_tools.push(call.clone());
sync_assistant_turn(
&messages_sent,
&full_reasoning,
reasoning_signature.as_deref(),
&full_text,
&proposed_tools,
&mut assistant_history_index,
);
let history = messages_sent
.lock()
.map(|sent| sent.clone())
.unwrap_or_default();
match tool_policy.decide(&call, &history, &tool_result_archive) {
ToolCallDecision::Execute => {
match build_tool_proposed(&task_id, &call, &skill_path_origin) {
Ok(action) => yield Ok(StreamEvent::ToolProposed(action)),
Err(message) => {
yield Err(agent_error(AgentError::new(
galaxy_agent_core::AgentErrorKind::Protocol,
message,
), stream_type));
return;
}
}
}
ToolCallDecision::Inline(result) => {
append_tool_result(&messages_sent, result);
}
ToolCallDecision::Reject(result) => {
log::warn!(
"Rig model called unavailable tool '{}' (id={})",
call.name,
call.id
);
let error_display = format!(
"Failed tool call: `{}`\n\n{}",
call.name, result.content
);
append_tool_result(&messages_sent, result);
let message_id = Uuid::new_v4().to_string();
yield Ok(StreamEvent::Response(build_add_agent_output_message(
&task_id,
&message_id,
&error_display,
)));
}
}
}
AgentEvent::TurnStopped { reason } => {
if !initialized {
yield Ok(StreamEvent::Response(build_stream_init(&request_id, &conversation_id)));
}
sync_assistant_turn(
&messages_sent,
&full_reasoning,
reasoning_signature.as_deref(),
&full_text,
&proposed_tools,
&mut assistant_history_index,
);
yield Ok(StreamEvent::Response(build_stream_finished(
map_stop_reason(reason),
StreamUsage {
input_tokens: saturating_i32(usage.input_tokens),
output_tokens: saturating_i32(usage.output_tokens),
cache_read_tokens: saturating_i32(usage.cached_input_tokens),
cache_write_tokens: saturating_i32(
usage.cache_creation_input_tokens,
),
cost_in_cents: 0.0,
model_id,
max_context_tokens,
},
)));
return;
}
AgentEvent::Tool { .. } => {
yield Err(agent_error(AgentError::new(
galaxy_agent_core::AgentErrorKind::Protocol,
"the provider runtime attempted to execute a tool outside Galaxy's permission boundary",
), stream_type));
return;
}
AgentEvent::RuntimeActivityUpdated { .. }
| AgentEvent::ContextUsageUpdated { .. }
| AgentEvent::UserInputAccepted { .. }
| AgentEvent::RuntimeNotice { .. } => {
yield Err(agent_error(AgentError::new(
galaxy_agent_core::AgentErrorKind::Protocol,
"the provider runtime emitted a session-runtime event",
), stream_type));
return;
}
}
}
}
}
};
Box::pin(stream)
}
fn store_messages_sent(
messages_sent: &std::sync::Arc<std::sync::Mutex<Vec<ConversationMessage>>>,
messages: &[ConversationMessage],
) {
let Ok(mut sent) = messages_sent.lock() else {
return;
};
*sent = messages.to_vec();
}
fn append_tool_result(
messages_sent: &std::sync::Arc<std::sync::Mutex<Vec<ConversationMessage>>>,
result: ToolResult,
) {
let is_error = result.is_error();
let message = ConversationMessage {
role: MessageRole::User,
content: MessageContent::ToolResult {
tool_use_id: result.call_id,
content: result.content,
is_error,
},
};
if let Ok(mut sent) = messages_sent.lock() {
sent.push(message);
}
}
fn sync_assistant_turn(
messages_sent: &std::sync::Arc<std::sync::Mutex<Vec<ConversationMessage>>>,
reasoning_text: &str,
reasoning_signature: Option<&str>,
text: &str,
tool_calls: &[ToolCall],
history_index: &mut Option<usize>,
) {
let has_reasoning = !reasoning_text.is_empty() || reasoning_signature.is_some();
let mut parts = Vec::with_capacity(
usize::from(has_reasoning) + usize::from(!text.is_empty()) + tool_calls.len(),
);
if has_reasoning {
parts.push(ContentPart::Reasoning {
text: reasoning_text.to_string(),
signature: reasoning_signature.map(str::to_string),
});
}
if !text.is_empty() {
parts.push(ContentPart::Text(text.to_string()));
}
parts.extend(tool_calls.iter().map(|call| ContentPart::ToolUse {
tool_use_id: call.id.clone(),
name: call.name.clone(),
input: call.arguments.clone(),
}));
if parts.is_empty() {
return;
}
let content = if parts.len() == 1 {
match parts.pop().unwrap() {
ContentPart::Text(text) => MessageContent::Text(text),
ContentPart::ToolUse {
tool_use_id,
name,
input,
} => MessageContent::ToolUse {
tool_use_id,
name,
input,
},
reasoning @ ContentPart::Reasoning { .. } => MessageContent::MultiPart(vec![reasoning]),
ContentPart::Image { .. } | ContentPart::ToolResult { .. } => unreachable!(),
}
} else {
MessageContent::MultiPart(parts)
};
let message = ConversationMessage {
role: MessageRole::Assistant,
content,
};
let Ok(mut sent) = messages_sent.lock() else {
return;
};
if let Some(index) = *history_index {
if index < sent.len() {
sent[index] = message;
return;
}
}
*history_index = Some(sent.len());
sent.push(message);
}
fn build_tool_proposed(
task_id: &str,
call: &ToolCall,
skill_path_origin: &ai::skills::SkillPathOrigin,
) -> Result<AIAgentAction, String> {
action_from_tool_call(task_id, call, skill_path_origin)
}
fn build_add_reasoning(task_id: &str, message_id: &str, text: &str) -> ResponseEvent {
reasoning_action(task_id, message_id, text, false)
}
fn build_append_reasoning(task_id: &str, message_id: &str, text: &str) -> ResponseEvent {
reasoning_action(task_id, message_id, text, true)
}
fn reasoning_action(task_id: &str, message_id: &str, text: &str, append: bool) -> ResponseEvent {
let message = api::Message {
id: message_id.to_string(),
task_id: task_id.to_string(),
request_id: String::new(),
timestamp: None,
server_message_data: String::new(),
citations: Vec::new(),
fetched_memories: Vec::new(),
message: Some(api::message::Message::AgentReasoning(
api::message::AgentReasoning {
reasoning: text.to_string(),
finished_duration: None,
},
)),
};
let action = if append {
api::client_action::Action::AppendToMessageContent(
api::client_action::AppendToMessageContent {
task_id: task_id.to_string(),
message: Some(message),
mask: Some(prost_types::FieldMask {
paths: vec!["agent_reasoning.reasoning".to_string()],
}),
},
)
} else {
api::client_action::Action::AddMessagesToTask(api::client_action::AddMessagesToTask {
task_id: task_id.to_string(),
messages: vec![message],
})
};
ResponseEvent {
r#type: Some(api::response_event::Type::ClientActions(
api::response_event::ClientActions {
actions: vec![ClientAction {
action: Some(action),
}],
},
)),
}
}
fn map_stop_reason(reason: StopReason) -> stream_finished::Reason {
match reason {
StopReason::Completed => stream_finished::Reason::Done(stream_finished::Done {}),
StopReason::MaxTokens => {
stream_finished::Reason::MaxTokenLimit(stream_finished::ReachedMaxTokenLimit {})
}
StopReason::ContextWindowExceeded => stream_finished::Reason::ContextWindowExceeded(
stream_finished::ContextWindowExceeded {},
),
StopReason::Cancelled
| StopReason::Refusal
| StopReason::ToolLoopLimit
| StopReason::Other(_) => stream_finished::Reason::Other(stream_finished::Other {}),
}
}
fn saturating_i32(value: u64) -> i32 {
i32::try_from(value).unwrap_or(i32::MAX)
}
fn agent_error(error: AgentError, stream_type: &'static str) -> Arc<AIApiError> {
Arc::new(
AIApiError::Stream {
stream_type,
source: anyhow::anyhow!(error),
}
.into_quota_limit_if_provider_budget_exhausted(),
)
}
#[cfg(test)]
#[path = "rig_tests.rs"]
mod tests;