Configuration Reference
A user-focused guide to YeePilot's main settings: AI behavior, safety, sandboxing, files, vault, updates, and TUI controls.
Where The Config Lives
YeePilot writes its main config to:
~/.yeepilot/config.ymlLegacy note:
- old installs may still have
~/.yeepilot/config.yaml - YeePilot can still read that legacy file
- new saves use
config.yml
Reset and regenerate the config through the setup flow:
yeepilot setup --resetA Good Starting Example
ai:
provider: openrouter
model: openai/gpt-oss-120b:free
autonomy_profile: guarded
vibe_coding: true
token_mode: normal
think_mode: medium
security:
mode: strict
require_confirmation: true
sandbox:
enabled: true
use_namespaces: true
network_access: true
tui:
theme: auto
live_stream_panel_enabled: true
vault:
enabled: false
auto_lock_duration: 15m
update:
auto_check: true
channel: publicai: Provider, Model, And Runtime Behavior
These are the settings most users will actually touch:
| Key | What it does |
|---|---|
ai.provider | Chooses openrouter, openai, or anthropic |
ai.model | Sets the default model |
ai.api_key | Stores the provider credential if you are not using env vars or keyring storage |
ai.base_url | Overrides the API endpoint |
ai.max_tokens | Caps response size |
ai.token_mode | normal or saver |
ai.conversation_max_history | Limits how much recent conversation is kept verbatim |
ai.output_truncate_length | Limits how much command output is fed back into the model |
ai.max_response_length | Sets a hard response-size cap |
ai.engine | Usually keep v2 |
ai.autonomy_profile | high, guarded, strict_review, or ultra |
ai.vibe_coding | Local on/off switch for coding-agent capability when your plan allows it |
ai.think_mode | off, auto, best, low, medium, or high |
Provider defaults
If you set only ai.provider, YeePilot fills in the default model and base URL automatically:
- OpenRouter:
openai/gpt-oss-120b:free - OpenAI:
gpt-4o - Anthropic:
claude-sonnet-4-20250514
ai.codeintel: Better Local Context For Repos
These settings improve how YeePilot understands local projects:
ai:
codeintel:
enabled: true
index_backend: sqlite
ast_enabled: true
embeddings_enabled: true
embeddings_provider: provider
max_indexed_files: 10000
max_file_size_bytes: 524288
rebuild_on_startup: falseMost end users should leave these defaults alone unless indexing needs to be smaller or faster.
ai.verification, ai.recovery, And ai.agentic
These sections control how carefully YeePilot verifies work and how much automatic repair it attempts.
Verification
ai:
verification:
require_post_mutation_verify: true
max_verify_retries: 1Recovery
ai:
recovery:
enabled: true
global_max_cycles: 4
auto_fix_cycles_max: 4
stop_on_repeated_signature: true
require_verify_after_auto_fix: true
missing_dependency_policy: smartmissing_dependency_policy values:
smartfallback_onlyauto_install
For most end users, smart is the safest balance.
Agentic behavior
ai:
agentic:
preedit_auto_context_enabled: true
recovery_auto_context_enabled: true
auto_verify_rerun_enabled: true
auto_checklist_rerun_enabled: true
auto_patch_synthesis_enabled: true
auto_patch_max_candidates: 2
auto_patch_apply_strategy: patch_first
auto_patch_confidence_threshold: 0.70
max_auto_repair_passes: 3These are advanced controls. End users normally leave them at their defaults.
ai.knowledge And ai.context7
These settings influence freshness and external knowledge retrieval:
ai:
knowledge:
enabled: true
freshness_policy: auto_live
offline_behavior: graceful_fallback
source_policy: official_first
max_sources: 4
cache_ttl: 6h
context7:
enabled: true
base_url: https://context7.com
max_chars: 8000Recommended for normal end-user use:
- keep
knowledge.enabled: true - keep
offline_behavior: graceful_fallback - keep
source_policy: official_first
ai.failover
If you want fallback providers:
ai:
failover:
enabled: true
chain:
- provider: openai
model: gpt-4o
- provider: openrouter
model: openai/gpt-oss-120b:freeThis is optional, but useful if provider uptime matters to you.
security
security:
mode: strict
require_confirmation: true
blocked_patterns:
- "shutdown"
audit_log_path: ~/.yeepilot/audit.log
audit_max_size_mb: 10
audit_max_archives: 50Key end-user advice:
- use
stricton production or important machines - keep
require_confirmation: true - use
blocked_patternsfor your own environment-specific no-go commands
sandbox
sandbox:
enabled: true
use_namespaces: true
network_access: true
max_cpu_seconds: 300
max_memory_mb: 512
max_filesize_mb: 100
max_processes: 64
denied_paths:
- /etc/shadow
- /etc/gshadow
- /rootThis is the main execution isolation section. Most users should leave it enabled and only tune network access or limits when a task clearly needs it.
tui
These are the main user-facing interface settings:
tui:
theme: auto
sidecar_emoji_mode: off
stats_panel: true
sidecar_module: sysmon
code_line_numbers: false
live_stream_panel_enabled: true
live_stream_panel_height: 12
live_stream_panel_scope: all_mutating_tools
live_stream_panel_animation: true
live_stream_panel_redact_sensitive: trueUseful values:
theme:auto,dark,light,neonsidecar_module:sysmon,vault,tasks,notes,explorerlive_stream_panel_scope:all_mutating_tools,file_tools_only,off
About HUD and keybind config
You may also see:
tui.hud.*tui.keybinds.actions
These are real settings, but most end users should manage them through the Neural HUD instead of editing them manually.
files
files:
max_entries: 20
max_file_size: 102400
max_total: 524288
upload_dir: ~/.yeepilot/uploads
clipboard_paste_mode: autoclipboard_paste_mode values:
autoonoff
vault
vault:
enabled: false
start_locked: true
auto_lock_duration: 15m
bruteforce_protection_enabled: true
wipe_on_bruteforce_enabled: false
wipe_after_failures: 15Important note:
wipe_on_bruteforce_enabledis intentionally high risk- only enable it if you fully understand the recovery consequences
personality
personality:
tone: casual
language: auto
use_emoji: false
verbosity: normalUseful values:
tone:casual,professional,minimallanguage:auto,en,deverbosity:normal,brief,detailed
platform
platform:
server_url: https://yee.toThis is the base URL for platform auth, heartbeats, diagnostics, and entitlement-aware client calls.
update
update:
auto_check: true
check_interval_hours: 1
channel: public
channel_token: ""Useful values:
channel:publicorcanarychannel_token: required for protected canary usage
Best End-User Editing Strategy
Use these three methods in this order:
- the setup wizard for provider and auth changes
- the Neural HUD Settings tab for day-to-day adjustments
- direct editing of
config.ymlfor advanced or bulk changes
That keeps the configuration understandable and avoids accidental low-level changes.