chore: remove autopilot experimental feature (#5781)
This commit is contained in:
Michael Neale
@@ -56,7 +56,6 @@ use tokio_util::sync::CancellationToken;
|
||||
use tracing::{debug, error, info, instrument, warn};
|
||||
|
||||
use super::final_output_tool::FinalOutputTool;
|
||||
use super::model_selector::autopilot::AutoPilot;
|
||||
use super::platform_tools;
|
||||
use super::tool_execution::{ToolCallResult, CHAT_MODE_TOOL_SKIPPED_RESPONSE, DECLINED_RESPONSE};
|
||||
use crate::agents::subagent_task_config::TaskConfig;
|
||||
@@ -105,7 +104,6 @@ pub struct Agent {
|
||||
pub(super) scheduler_service: Mutex<Option<Arc<dyn SchedulerTrait>>>,
|
||||
pub(super) retry_manager: RetryManager,
|
||||
pub(super) tool_inspection_manager: ToolInspectionManager,
|
||||
pub(super) autopilot: Mutex<AutoPilot>,
|
||||
}
|
||||
|
||||
#[derive(Clone, Debug)]
|
||||
@@ -180,7 +178,6 @@ impl Agent {
|
||||
scheduler_service: Mutex::new(None),
|
||||
retry_manager: RetryManager::new(),
|
||||
tool_inspection_manager: Self::create_default_tool_inspection_manager(),
|
||||
autopilot: Mutex::new(AutoPilot::new()),
|
||||
}
|
||||
}
|
||||
|
||||
@@ -933,19 +930,6 @@ impl Agent {
|
||||
break;
|
||||
}
|
||||
|
||||
{
|
||||
let mut autopilot = self.autopilot.lock().await;
|
||||
if let Some((new_provider, role, model)) = autopilot.check_for_switch(&conversation, self.provider().await?).await? {
|
||||
debug!("AutoPilot switching to {} role with model {}", role, model);
|
||||
self.update_provider(new_provider).await?;
|
||||
|
||||
yield AgentEvent::ModelChange {
|
||||
model: model.clone(),
|
||||
mode: format!("autopilot:{}", role),
|
||||
};
|
||||
}
|
||||
}
|
||||
|
||||
let conversation_with_moim = super::moim::inject_moim(
|
||||
conversation.clone(),
|
||||
&self.extension_manager,
|
||||
|
||||
@@ -7,7 +7,6 @@ pub mod extension_manager_extension;
|
||||
pub mod final_output_tool;
|
||||
mod large_response_handler;
|
||||
pub mod mcp_client;
|
||||
pub mod model_selector;
|
||||
pub mod moim;
|
||||
pub mod platform_tools;
|
||||
pub mod prompt_manager;
|
||||
|
||||
@@ -1,28 +0,0 @@
|
||||
# Autopilot model selector
|
||||
|
||||
This is an advanced feature (config of which may change, use with caution for now)
|
||||
which lets goose automatically rotate through many providers and models based on rules that trigger as part of its work.
|
||||
|
||||
Models can change at any time, and can help (similar to lead/worker) solve persistent issues, get an advanced plan, a second opinion or more.
|
||||
|
||||
`premade_roles.yaml` are the out of the box configurations, which can be used in the `~/.config/goose/config.yaml` like so:
|
||||
|
||||
|
||||
```yaml
|
||||
x-advanced-models:
|
||||
- provider: databricks
|
||||
model: goose-gpt-5
|
||||
role: reviewer
|
||||
- provider: anthropic
|
||||
model: claude-opus-4-1-20250805
|
||||
role: deep-thinker
|
||||
```
|
||||
|
||||
in this case, when there is some complex activity or planning or thining required, it will automatically switch to opus for a while, likewise when code changes have been made, it will use the reviewer model.
|
||||
|
||||
## Use cases
|
||||
|
||||
You can do a lead/worker like combo, or you can default to a low cost model and only in some cases use a frontier model.
|
||||
You could default to a local model, and only intermittently switch when needed.
|
||||
|
||||
use `--debug` flag if you want to see it logging when it changes.
|
||||
File diff suppressed because it is too large
Load Diff
@@ -1 +0,0 @@
|
||||
pub mod autopilot;
|
||||
@@ -1,181 +0,0 @@
|
||||
# Pre-made AutoPilot roles with default rules
|
||||
# These define the default behaviors for common roles
|
||||
# Users must specify the provider and model in their config.yaml
|
||||
|
||||
roles:
|
||||
# Lead model - high-capability model for initial turns and failure recovery
|
||||
- role: "lead"
|
||||
rules:
|
||||
triggers:
|
||||
# Triggers at conversation start AND on consecutive failures
|
||||
first_turn: true # Trigger on first turn
|
||||
consecutive_failures: 2 # Same as GOOSE_LEAD_FAILURE_THRESHOLD default
|
||||
source: "any" # Can trigger on both human (start) and machine (failures)
|
||||
active_turns: 3 # Same as GOOSE_LEAD_TURNS default (initial) and GOOSE_LEAD_FALLBACK_TURNS for failures
|
||||
priority: 30 # Highest priority to ensure it always triggers first
|
||||
|
||||
- role: "second-opinion"
|
||||
rules:
|
||||
triggers:
|
||||
keywords: ["help"]
|
||||
match_type: "any"
|
||||
after_tool_use: true
|
||||
source: "human"
|
||||
active_turns: 5
|
||||
priority: 5
|
||||
|
||||
# Deep reasoning and analysis
|
||||
- role: "deep-thinker"
|
||||
rules:
|
||||
triggers:
|
||||
keywords: ["think", "reason", "analyze", "explain why", "how does", "what if"]
|
||||
match_type: "any"
|
||||
complexity_threshold: "high"
|
||||
source: "human" # Only trigger on human messages
|
||||
active_turns: 3
|
||||
priority: 10
|
||||
|
||||
# Consult the oracle
|
||||
- role: "oracle"
|
||||
rules:
|
||||
triggers:
|
||||
keywords: ["think", "reason", "analyze", "explain why", "what if"]
|
||||
match_type: "any"
|
||||
complexity_threshold: "medium"
|
||||
source: "human" # Only trigger on human messages
|
||||
active_turns: 5
|
||||
priority: 15
|
||||
|
||||
# Consult the planner
|
||||
- role: "planner"
|
||||
rules:
|
||||
triggers:
|
||||
keywords: ["think", "plan", "help me", "look at", "consider"]
|
||||
match_type: "any"
|
||||
complexity_threshold: "low"
|
||||
source: "any" # Only trigger on human messages
|
||||
active_turns: 3
|
||||
priority: 5
|
||||
|
||||
|
||||
|
||||
# Code debugging and error recovery
|
||||
- role: "debugger"
|
||||
rules:
|
||||
triggers:
|
||||
keywords: ["error", "bug", "broken", "failed", "exception"]
|
||||
match_type: "any"
|
||||
on_failure: true
|
||||
source: "any" # Can trigger on both human and machine failures
|
||||
active_turns: 2
|
||||
priority: 15 # High priority for error handling
|
||||
|
||||
# Code implementation specialist
|
||||
- role: "coder"
|
||||
rules:
|
||||
triggers:
|
||||
keywords: ["implement", "code", "function", "class", "refactor", "optimize"]
|
||||
match_type: "any"
|
||||
after_tool_use: true
|
||||
source: "human"
|
||||
active_turns: 2
|
||||
priority: 8
|
||||
|
||||
# Verification and review
|
||||
- role: "reviewer"
|
||||
rules:
|
||||
triggers:
|
||||
keywords: ["check", "verify", "review", "validate", "test", "correct"]
|
||||
match_type: "any"
|
||||
consecutive_tools: 12 # After many changes
|
||||
source: "any" # Can be triggered by human request OR after lots of tool use
|
||||
active_turns: 2
|
||||
priority: 6
|
||||
|
||||
# Help and guidance specialist
|
||||
- role: "helper"
|
||||
rules:
|
||||
triggers:
|
||||
keywords: ["help", "assist", "guide", "explain", "teach", "how to"]
|
||||
match_type: "any"
|
||||
source: "human"
|
||||
active_turns: 5
|
||||
priority: 5
|
||||
|
||||
# Math and calculations
|
||||
- role: "mathematician"
|
||||
rules:
|
||||
triggers:
|
||||
keywords: ["calculate", "solve", "equation", "math", "formula", "compute"]
|
||||
match_type: "any"
|
||||
complexity_threshold: "medium"
|
||||
source: "human"
|
||||
active_turns: 1
|
||||
priority: 7
|
||||
|
||||
# Creative brainstorming
|
||||
- role: "creative"
|
||||
rules:
|
||||
triggers:
|
||||
keywords: ["idea", "brainstorm", "creative", "innovate", "design", "imagine"]
|
||||
match_type: "any"
|
||||
source: "human"
|
||||
active_turns: 5
|
||||
priority: 4
|
||||
|
||||
# Quick responses for simple queries
|
||||
- role: "quick-responder"
|
||||
rules:
|
||||
triggers:
|
||||
complexity_threshold: "low"
|
||||
source: "human"
|
||||
active_turns: 0
|
||||
priority: 2
|
||||
|
||||
# Research and fact-checking
|
||||
- role: "researcher"
|
||||
rules:
|
||||
triggers:
|
||||
keywords: ["research", "find", "search", "lookup", "fact", "source", "reference"]
|
||||
match_type: "any"
|
||||
source: "human"
|
||||
active_turns: 3
|
||||
priority: 6
|
||||
|
||||
# System recovery after multiple failures
|
||||
- role: "recovery-specialist"
|
||||
rules:
|
||||
triggers:
|
||||
consecutive_failures: 2 # After 2 consecutive tool failures
|
||||
source: "machine" # Only triggers on machine-generated failures
|
||||
active_turns: 10
|
||||
priority: 20 # Very high priority
|
||||
|
||||
# Autonomous work reviewer - kicks in after lots of machine work
|
||||
- role: "work-reviewer"
|
||||
rules:
|
||||
triggers:
|
||||
tools_since_human: 5 # After 5+ tools used since last human input
|
||||
source: "machine" # Only when machine is active
|
||||
active_turns: 8
|
||||
priority: 12
|
||||
|
||||
# Progress checker - ensures the machine isn't going off track
|
||||
- role: "progress-checker"
|
||||
rules:
|
||||
triggers:
|
||||
machine_messages_without_human: 4 # After 4+ consecutive machine messages
|
||||
source: "machine"
|
||||
active_turns: 5
|
||||
priority: 11
|
||||
|
||||
# Intensive work monitor - for when lots of tool use is happening
|
||||
- role: "intensive-work-monitor"
|
||||
rules:
|
||||
triggers:
|
||||
consecutive_tools: 10 # 10+ tools in a row
|
||||
messages_since_human: 6 # AND been working for 6+ messages
|
||||
source: "machine"
|
||||
active_turns: 10
|
||||
priority: 14
|
||||
|
||||
Reference in New Issue
Block a user