Configuration keys (generated)
Configuration keys (generated)
Section titled “Configuration keys (generated)”Intended readers: Operators, system administrators, and developers tuning Vox runtime knobs and pipeline limits.
Machine-derived from contracts/config/registry.v1.yaml.
Total knobs: 151
| Key | Environment Variable | Owner Crate | Status | Default | Description |
|---|---|---|---|---|---|
allow_main_push | VOX_ALLOW_MAIN_PUSH | vox-cli | active | — | Emergency override that lets the pre-push hook allow a direct push to main. |
broker_home | VOX_BROKER_HOME | vox-cli | active | — | Install root of the cargo build-broker shim (checked by vox doctor); defaults under the user home. |
broker_max_concurrent | VOX_BROKER_MAX_CONCURRENT | vox-build-queue | active | — | Optional machine-wide upper bound on simultaneous broker-admitted cargo builds; unset derives from the machine. |
browser_allowed_hosts | VOX_BROWSER_ALLOWED_HOSTS | vox-plugin-browser | active | — | Optional comma-separated browser navigation host allowlist (exact hosts or *.example.com wildcards). |
browser_no_sandbox | VOX_BROWSER_NO_SANDBOX | vox-plugin-browser | active | 0 | Launch Chromium without its sandbox (“1”/“true”); for containers that cannot run it. |
browser_profiles_dir | VOX_BROWSER_PROFILES_DIR | vox-config | active | — | Directory for named Chromium profiles and consents.json; default <data_dir>/browser-profiles. |
build_number | VOX_BUILD_NUMBER | vox-build-meta | active | — | Explicit build number stamped by vox-build-meta (release builds from shallow clones); unset derives from git. |
cache_dir | VOX_CACHE_DIR | vox-config | active | — | Override for the Vox cache directory; default is the platform cache dir. |
capability_profile | VOX_PROFILE | vox-mcp-registry | active | dev | Comma-separated capability profile ids scoping MCP/HTTP tools and daemon boot; unset or empty means dev. |
circuit_breaker_contract | VOX_CIRCUIT_BREAKER_CONTRACT | vox-orchestrator | active | — | Override path for the circuit-breaker contract YAML (falls back to compiled-in default). |
cost_daily_budget_usd | VOX_COST_DAILY_BUDGET_USD | vox-scaling-policy | declared | 25.0 | Daily LLM cost budget in USD. No live consumer yet — wired when vox-scaling-policy gains a consumer. |
cost_monthly_budget_usd | VOX_COST_MONTHLY_BUDGET_USD | vox-scaling-policy | declared | 500.0 | Monthly LLM cost budget in USD. No live consumer yet. |
db_path | VOX_DB_PATH | vox-db | active | — | Local database file path override for the Vox store. |
embedding_search_candidate_mult | VOX_EMBEDDING_SEARCH_CANDIDATE_MULT | vox-db | active | 10 | Multiplier on the result limit for brute-force embedding recall candidates (boosted when FTS5 is absent). |
gamify_economy_path | VOX_GAMIFY_ECONOMY_PATH | vox-gamify | active | — | Override path for the gamify economy contract YAML (falls back to compiled-in default). |
gui_drive | VOX_GUI_DRIVE | vox-gui | active | false | Marks the dedicated Axis Drive process (never attach to the user’s window). |
gui_drive_profile | VOX_GUI_DRIVE_PROFILE | vox-gui | active | default | Isolates ~/.vox/gui-drive/ |
gui_drive_session_path | VOX_GUI_DRIVE_SESSION_PATH | vox-gui | active | — | Path the drive child writes for gui-drive.json. |
gui_drive_show | VOX_GUI_DRIVE_SHOW | vox-gui | active | false | Reveal the dedicated debug Axis (hidden + no Dock icon by default). |
gui_drive_store_root | VOX_GUI_DRIVE_STORE_ROOT | vox-gui | active | — | Journey-store root so drive chat_turn cannot write the user transcript. |
gui_drive_token_path | VOX_GUI_DRIVE_TOKEN_PATH | vox-gui | active | — | Path to the 0600 bearer file. Never put the token value in the environment. |
gui_skip_sidecar_autobuild | VOX_GUI_SKIP_SIDECAR_AUTOBUILD | vox-gui | active | — | vox-gui build.rs does not auto-build missing sidecar binaries when set; it reports them as missing instead. |
llm_request_timeout_ms | — | vox-actor-runtime | active | 30000 | Per-request timeout for unary LLM HTTP calls (chat/embed). Set via LlmConfig.timeout_ms. |
local_endpoint | VOX_LOCAL_ENDPOINT | vox-config | active | — | Base URL of the VoxLocal (vox mens serve) inference server; unset probes 127.0.0.1:11434 then :11435. |
mcp_workspace_tools | VOX_MCP_WORKSPACE_TOOLS | vox-orchestrator-mcp | active | 0 | Advertise workspace-federated @tool functions in MCP tools/list (dev profile only). Replaces the deprecated VOX_MCP_TIERS opt-in. |
mens_allow_spend | VOX_MENS_ALLOW_SPEND | vox-ml-cli | active | — | MENS pipeline is allowed to make paid LLM calls (apply mode) when set to “1”; otherwise it only estimates spend. |
mens_base_only_dir | VOX_MENS_BASE_ONLY_DIR | vox-plugin-mens-candle-metal | declared | — | Test-only. Base-model directory (no adapter) for the ignored live-Metal base-only inference test. |
mens_default_model | VOX_MENS_DEFAULT_MODEL | vox-populi | active | Qwen/Qwen2.5-Coder-7B-Instruct | Default base model id for MENS training/inference. |
mens_demo_run_dir | VOX_MENS_DEMO_RUN_DIR | vox-plugin-mens-candle-metal | declared | — | Test-only. MENS run directory for the ignored live-Metal inference gates. |
mens_device | VOX_MENS_DEVICE | vox-cli | active | cuda | Override training device for full-pipeline.vox (e.g. metal for a Mac run). |
mens_force_train | VOX_MENS_FORCE_TRAIN | vox-ml-cli | active | 0 | MENS GPU training proceeds even when the VRAM sizing check says the run will not fit (“1”/“true”). |
mens_no_weight_cache | VOX_MENS_NO_WEIGHT_CACHE | vox-plugin-mens-candle-metal | active | — | Candle/Metal QLoRA training does not cache dequantized base weights when set (lower memory, slower steps). |
mens_preset | VOX_MENS_PRESET | vox-cli | active | qwen_4080_16g | Override training preset for full-pipeline.vox. |
mesh_mailbox_test_admission_pause_ms | MAILBOX_TEST_ADMISSION_PAUSE_MS | vox-mesh-transport | declared | — | Test-only (cfg(test)). Pause inside mailbox admission to widen the race window in capacity tests. |
mesh_mailbox_test_capacity_child | MAILBOX_TEST_CAPACITY_CHILD | vox-mesh-transport | declared | — | Test-only. Marks a child process in the mailbox capacity multi-process test. |
mesh_mailbox_test_child_index | MAILBOX_TEST_CHILD_INDEX | vox-mesh-transport | declared | — | Test-only. Index of a mailbox capacity test child process. |
mesh_mailbox_test_dir | MAILBOX_TEST_DIR | vox-mesh-transport | declared | — | Test-only. Shared mailbox directory handed to mailbox capacity test child processes. |
model_force | VOX_MODEL_FORCE | vox-config | active | — | Global strict model pin (env first, then ~/.vox/config.toml); a pinned model that is missing or gated errors. |
openrouter_base_url | OPENROUTER_BASE_URL | vox-gamify | active | — | Override the OpenRouter API base URL (test/mock server injection point in vox-gamify and vox-orchestrator’s OpenRouter clients). Follows OpenRouter’s own env-var naming, not the VOX_ prefix convention. |
oratio_domain_mode | VOX_ORATIO_DOMAIN_MODE | vox-speech | active | — | Speech-to-text domain post-processing mode (vocabulary correction profile) for Oratio transcription. |
oratio_sherpa_kind | VOX_ORATIO_SHERPA_KIND | vox-speech | active | — | Shape of the sherpa-onnx model at VOX_ORATIO_SHERPA_MODEL_DIR (“whisper” selects the Whisper loader). |
orchestrator_agentos_aci_envelope_enabled | VOX_ORCHESTRATOR_AGENTOS_ACI_ENVELOPE_ENABLED | vox-orchestrator | active | true | Wrap agent completions in the ACI (Agent Completion Interface) envelope format. |
orchestrator_agentos_checkpoint_hints_enabled | VOX_ORCHESTRATOR_AGENTOS_CHECKPOINT_HINTS_ENABLED | vox-orchestrator | active | false | Emit checkpoint hints that agents may use to save and resume long-running tasks. |
orchestrator_agentos_guardrail_kernel_enabled | VOX_ORCHESTRATOR_AGENTOS_GUARDRAIL_KERNEL_ENABLED | vox-orchestrator | active | false | Enable the AgentOS guardrail kernel for runtime policy enforcement. |
orchestrator_attention_alert_threshold | VOX_ORCHESTRATOR_ATTENTION_ALERT_THRESHOLD | vox-orchestrator | active | 0.7 | Fraction of budget consumed (0.0–1.0) that triggers an AttentionHigh signal. |
orchestrator_attention_budget_ms | VOX_ORCHESTRATOR_ATTENTION_BUDGET_MS | vox-orchestrator | active | 3600000 | Per-session attention budget in ms (1 hr default). |
orchestrator_attention_enabled | VOX_ORCHESTRATOR_ATTENTION_ENABLED | vox-orchestrator | active | true | Enable the attention-budget subsystem for session cost accounting. |
orchestrator_attention_interrupt_cost_ms | VOX_ORCHESTRATOR_ATTENTION_INTERRUPT_COST_MS | vox-orchestrator | active | 23250 | Estimated attention cost in ms charged per mid-session interrupt. |
orchestrator_attention_trust_routing_weight | VOX_ORCHESTRATOR_ATTENTION_TRUST_ROUTING_WEIGHT | vox-orchestrator | active | 2.0 | Weight given to attention-trust score when selecting agents for routing. |
orchestrator_chatml_strict | VOX_ORCHESTRATOR_CHATML_STRICT | vox-orchestrator | active | false | Enforce strict ChatML message formatting; reject malformed role sequences. |
orchestrator_completion_grounding_enforce | VOX_ORCHESTRATOR_COMPLETION_GROUNDING_ENFORCE | vox-orchestrator | active | false | Enforce completion-grounding checks (block ungrounded completions). |
orchestrator_completion_grounding_shadow | VOX_ORCHESTRATOR_COMPLETION_GROUNDING_SHADOW | vox-orchestrator | active | false | Run completion-grounding checks in shadow mode. |
orchestrator_completion_markdown_link_audit_enabled | VOX_ORCHESTRATOR_COMPLETION_MARKDOWN_LINK_AUDIT_ENABLED | vox-orchestrator | active | true | Audit markdown links in agent completions for broken/unresolved references. |
orchestrator_context_lifecycle_enforce | VOX_ORCHESTRATOR_CONTEXT_LIFECYCLE_ENFORCE | vox-orchestrator | active | false | Enforce context-lifecycle policy (reject out-of-lifecycle operations). |
orchestrator_context_lifecycle_shadow | VOX_ORCHESTRATOR_CONTEXT_LIFECYCLE_SHADOW | vox-orchestrator | active | false | Run context-lifecycle enforcement in shadow/observe mode. |
orchestrator_cost_preference | VOX_ORCHESTRATOR_COST_PREFERENCE | vox-orchestrator | active | economy | Model-selection cost preference: economy (default) or performance. |
orchestrator_enabled | VOX_ORCHESTRATOR_ENABLED | vox-orchestrator | active | true | Master on/off switch for the multi-agent orchestrator. |
orchestrator_exec_time_budget_enabled | VOX_ORCHESTRATOR_EXEC_TIME_BUDGET_ENABLED | vox-orchestrator | active | true | Enable per-tool execution-time budget learning and enforcement. |
orchestrator_exec_time_default_budget_ms | VOX_ORCHESTRATOR_EXEC_TIME_DEFAULT_BUDGET_MS | vox-orchestrator | active | 30000 | Default execution-time budget (ms) used when no historical data is available. |
orchestrator_exec_time_history_window_days | VOX_ORCHESTRATOR_EXEC_TIME_HISTORY_WINDOW_DAYS | vox-orchestrator | active | 30 | Number of days of execution-time history retained for budget calibration. |
orchestrator_exec_time_safety_multiplier | VOX_ORCHESTRATOR_EXEC_TIME_SAFETY_MULTIPLIER | vox-orchestrator | active | 2.0 | Safety multiplier applied to the P90 execution time when deriving the recommended budget. |
orchestrator_exec_time_timeout_rate_alert | VOX_ORCHESTRATOR_EXEC_TIME_TIMEOUT_RATE_ALERT | vox-orchestrator | active | 0.20 | Fraction of tool invocations that time out before an alert is raised. |
orchestrator_fallback_single | VOX_ORCHESTRATOR_FALLBACK_SINGLE | vox-orchestrator | active | true | Fall back to single-agent mode when multi-agent orchestration fails to start. |
orchestrator_idle_retirement_ms | VOX_ORCHESTRATOR_IDLE_RETIREMENT_MS | vox-orchestrator | active | 300000 | How long (ms) an idle agent lives before being retired by the scaler. |
orchestrator_interruption_cal_a2a_gain | VOX_ORCHESTRATOR_INTERRUPTION_CAL_A2A_GAIN | vox-orchestrator | active | 0.0 | Gain offset (bits) added to the A2A-escalation interruption benefit signal. |
orchestrator_interruption_cal_backlog_penalty | VOX_ORCHESTRATOR_INTERRUPTION_CAL_BACKLOG_PENALTY | vox-orchestrator | active | 0.0 | Cost penalty per backlog item applied to interruption cost modelling. |
orchestrator_interruption_cal_plan_gain | VOX_ORCHESTRATOR_INTERRUPTION_CAL_PLAN_GAIN | vox-orchestrator | active | 0.0 | Gain offset (bits) added to the plan-review interruption benefit signal. |
orchestrator_lock_timeout_ms | VOX_ORCHESTRATOR_LOCK_TIMEOUT_MS | vox-orchestrator | active | 30000 | Timeout in ms before an orchestrator state lock is considered stale and forcibly released. |
orchestrator_max_agents | VOX_ORCHESTRATOR_MAX_AGENTS | vox-orchestrator | active | 8 | Hard ceiling on concurrent orchestrated agents. |
orchestrator_max_debug_iterations | VOX_ORCHESTRATOR_MAX_DEBUG_ITERATIONS | vox-orchestrator | active | 5 | Maximum debug-iteration loops before the orchestrator aborts a stuck agent. |
orchestrator_max_socrates_debug_iterations | VOX_ORCHESTRATOR_MAX_SOCRATES_DEBUG_ITERATIONS | vox-orchestrator | active | 3 | Maximum Socrates-driven debug iterations before the gate gives up. |
orchestrator_max_spawn_per_tick | VOX_ORCHESTRATOR_MAX_SPAWN_PER_TICK | vox-orchestrator | active | 5 | Maximum number of new agents that may be spawned in a single scaling tick. |
orchestrator_max_toestub_debug_iterations | VOX_ORCHESTRATOR_MAX_TOESTUB_DEBUG_ITERATIONS | vox-orchestrator | active | 3 | Maximum toestub debug-loop iterations before the orchestrator aborts. |
orchestrator_mesh_control_url | VOX_ORCHESTRATOR_MESH_CONTROL_URL | vox-orchestrator | active | — | Override URL for the Populi mesh control plane API. |
orchestrator_mesh_http_timeout_ms | VOX_ORCHESTRATOR_MESH_HTTP_TIMEOUT_MS | vox-orchestrator | active | 10000 | HTTP timeout (ms) for mesh federation requests. |
orchestrator_mesh_poll_interval_secs | VOX_ORCHESTRATOR_MESH_POLL_INTERVAL_SECS | vox-orchestrator | active | 30 | Interval (s) between Populi mesh control-plane polls. |
orchestrator_mesh_rebalance_on_remote_schedulable_drop | VOX_ORCHESTRATOR_MESH_REBALANCE_ON_REMOTE_SCHEDULABLE_DROP | vox-orchestrator | active | false | Trigger a local rebalance when a remote schedulable agent drops from the mesh. |
orchestrator_mesh_remote_execute_experimental | VOX_ORCHESTRATOR_MESH_REMOTE_EXECUTE_EXPERIMENTAL | vox-orchestrator | active | false | Enable experimental remote-execution via the mesh. |
orchestrator_mesh_remote_lease_gating_enabled | VOX_ORCHESTRATOR_MESH_REMOTE_LEASE_GATING_ENABLED | vox-orchestrator | active | false | Gate remote task execution behind lease acquisition to prevent double-dispatch. |
orchestrator_mesh_remote_result_max_messages_per_poll | VOX_ORCHESTRATOR_MESH_REMOTE_RESULT_MAX_MESSAGES_PER_POLL | vox-orchestrator | active | 64 | Maximum result messages consumed per remote-result poll cycle. |
orchestrator_mesh_remote_result_poll_interval_secs | VOX_ORCHESTRATOR_MESH_REMOTE_RESULT_POLL_INTERVAL_SECS | vox-orchestrator | active | 5 | Interval (s) between polls for remote-execution results. |
orchestrator_mesh_remote_worker_poll_interval_secs | VOX_ORCHESTRATOR_MESH_REMOTE_WORKER_POLL_INTERVAL_SECS | vox-orchestrator | active | 5 | Interval (s) between polls for available remote worker capacity. |
orchestrator_mesh_replay_queued_routes_on_remote_schedulable_drop | VOX_ORCHESTRATOR_MESH_REPLAY_QUEUED_ROUTES_ON_REMOTE_SCHEDULABLE_DROP | vox-orchestrator | active | false | Replay queued routes when a remote schedulable agent drops from the mesh. |
orchestrator_mesh_routing_experimental | VOX_ORCHESTRATOR_MESH_ROUTING_EXPERIMENTAL | vox-orchestrator | active | false | Enable experimental mesh-based task routing. |
orchestrator_mesh_training_budget_pressure | VOX_ORCHESTRATOR_MESH_TRAINING_BUDGET_PRESSURE | vox-orchestrator | active | 0.0 | Soft budget-pressure scalar (0.0–1.0) applied to expensive training placements. |
orchestrator_mesh_training_routing_experimental | VOX_ORCHESTRATOR_MESH_TRAINING_ROUTING_EXPERIMENTAL | vox-orchestrator | active | false | Enable experimental mesh routing for training tasks. |
orchestrator_migration_legacy_fallback | VOX_ORCHESTRATOR_MIGRATION_LEGACY_FALLBACK | vox-orchestrator | active | true | Allow fallback to the legacy orchestration path when v2 errors. |
orchestrator_migration_v2_enabled | VOX_ORCHESTRATOR_MIGRATION_V2_ENABLED | vox-orchestrator | active | false | Enable the v2 orchestration code path (migration gate). |
orchestrator_min_agents | VOX_ORCHESTRATOR_MIN_AGENTS | vox-orchestrator | active | 1 | Minimum number of live agents before scale-down is blocked. |
orchestrator_plan_adequacy_enforce | VOX_ORCHESTRATOR_PLAN_ADEQUACY_ENFORCE | vox-orchestrator | active | false | Enforce the plan-adequacy gate (block execution on low-adequacy plans). |
orchestrator_plan_adequacy_shadow | VOX_ORCHESTRATOR_PLAN_ADEQUACY_SHADOW | vox-orchestrator | active | false | Run the plan-adequacy gate in shadow mode (evaluate but never block). |
orchestrator_planning_auto_mode_enabled | VOX_ORCHESTRATOR_PLANNING_AUTO_MODE_ENABLED | vox-orchestrator | active | false | Allow the orchestrator to autonomously select planning depth. |
orchestrator_planning_enabled | VOX_ORCHESTRATOR_PLANNING_ENABLED | vox-orchestrator | active | false | Enable the task-planning subsystem inside the orchestrator. |
orchestrator_planning_replan_enabled | VOX_ORCHESTRATOR_PLANNING_REPLAN_ENABLED | vox-orchestrator | active | false | Allow the orchestrator to re-plan mid-task when progress stalls. |
orchestrator_planning_rollout_percent | VOX_ORCHESTRATOR_PLANNING_ROLLOUT_PERCENT | vox-orchestrator | active | 0 | Percentage (0–100) of sessions that use the planning subsystem during gradual rollout. |
orchestrator_planning_router_enabled | VOX_ORCHESTRATOR_PLANNING_ROUTER_ENABLED | vox-orchestrator | active | false | Route tasks through the planning layer before agent dispatch. |
orchestrator_planning_shadow_mode | VOX_ORCHESTRATOR_PLANNING_SHADOW_MODE | vox-orchestrator | active | false | Run the planner in shadow mode (decisions logged but not applied). |
orchestrator_planning_workflow_handoff_enabled | VOX_ORCHESTRATOR_PLANNING_WORKFLOW_HANDOFF_ENABLED | vox-orchestrator | active | false | Enable workflow-level handoff signals between planning stages. |
orchestrator_populi_inference_base_url | VOX_ORCHESTRATOR_POPULI_INFERENCE_BASE_URL | vox-orchestrator | active | — | Override base URL for Populi mesh inference requests. |
orchestrator_repo_reduce_conflict_cooldown_ms | VOX_ORCHESTRATOR_REPO_REDUCE_CONFLICT_COOLDOWN_MS | vox-orchestrator | active | 300000 | Duration (ms) of the reducer conflict cooldown window. |
orchestrator_repo_reduce_conflict_cooldown_penalty | VOX_ORCHESTRATOR_REPO_REDUCE_CONFLICT_COOLDOWN_PENALTY | vox-orchestrator | active | 2.5 | Penalty applied to reducer placement while an agent is in conflict-cooldown. |
orchestrator_repo_shard_specialization_weight | VOX_ORCHESTRATOR_REPO_SHARD_SPECIALIZATION_WEIGHT | vox-orchestrator | active | 1.5 | Routing bonus weight for agents specialised in the target repo shard. |
orchestrator_repo_shard_validation_failure_penalty | VOX_ORCHESTRATOR_REPO_SHARD_VALIDATION_FAILURE_PENALTY | vox-orchestrator | active | 0.8 | Penalty multiplier applied per recent shard-validation failure when routing. |
orchestrator_research_model_enabled | VOX_ORCHESTRATOR_RESEARCH_MODEL_ENABLED | vox-orchestrator | active | false | Enable the dedicated research-model routing path inside the orchestrator. |
orchestrator_resource_cpu_mult | VOX_ORCHESTRATOR_RESOURCE_CPU_MULT | vox-orchestrator | active | 0.7 | Multiplier applied to the CPU-utilisation component of the resource score. |
orchestrator_resource_exponent | VOX_ORCHESTRATOR_RESOURCE_EXPONENT | vox-orchestrator | active | 1.0 | Exponent applied to the combined resource score (>1 super-linear, <1 sub-linear). |
orchestrator_resource_mem_mult | VOX_ORCHESTRATOR_RESOURCE_MEM_MULT | vox-orchestrator | active | 0.3 | Multiplier applied to the memory-utilisation component of the resource score. |
orchestrator_resource_weight | VOX_ORCHESTRATOR_RESOURCE_WEIGHT | vox-orchestrator | active | 0.3 | Weight applied to the composite resource-utilisation score in routing decisions. |
orchestrator_scaling_cooldown_ms | VOX_ORCHESTRATOR_SCALING_COOLDOWN_MS | vox-orchestrator | active | 5000 | Minimum interval (ms) between successive scale-up or scale-down decisions. |
orchestrator_scaling_enabled | VOX_ORCHESTRATOR_SCALING_ENABLED | vox-orchestrator | active | false | Enable dynamic agent pool scaling based on queue depth. |
orchestrator_scaling_lookback | VOX_ORCHESTRATOR_SCALING_LOOKBACK | vox-orchestrator | active | 5 | Number of ticks included in the rolling queue-depth window used for scale decisions. |
orchestrator_scaling_profile | VOX_ORCHESTRATOR_SCALING_PROFILE | vox-orchestrator | active | balanced | Scaling aggressiveness preset: conservative | balanced | aggressive. |
orchestrator_scaling_threshold | VOX_ORCHESTRATOR_SCALING_THRESHOLD | vox-orchestrator | active | 5 | Queue depth that triggers a scale-up decision. |
orchestrator_socrates_gate_enforce | VOX_ORCHESTRATOR_SOCRATES_GATE_ENFORCE | vox-orchestrator | active | false | Enforce Socrates quality gate (block low-quality completions). |
orchestrator_socrates_gate_shadow | VOX_ORCHESTRATOR_SOCRATES_GATE_SHADOW | vox-orchestrator | active | false | Run the Socrates quality gate in shadow mode (evaluate but never block). |
orchestrator_socrates_reputation_routing | VOX_ORCHESTRATOR_SOCRATES_REPUTATION_ROUTING | vox-orchestrator | active | false | Factor Socrates reputation scores into agent routing decisions. |
orchestrator_socrates_reputation_weight | VOX_ORCHESTRATOR_SOCRATES_REPUTATION_WEIGHT | vox-orchestrator | active | 1.0 | Weight of the Socrates reputation component in the routing score. |
orchestrator_tier_gate_entropy_threshold | VOX_ORCHESTRATOR_TIER_GATE_ENTROPY_THRESHOLD | vox-orchestrator | active | 1.0 | Entropy threshold below which the tier gate auto-approves without LLM review. |
orchestrator_tier_gate_min_observations | VOX_ORCHESTRATOR_TIER_GATE_MIN_OBSERVATIONS | vox-orchestrator | active | 10 | Minimum observations required before entropy-based auto-approve is eligible. |
orchestrator_tlx_frustration | VOX_ORCHESTRATOR_TLX_FRUSTRATION | vox-orchestrator | active | 1.0 | NASA-TLX frustration weight used in agent workload scoring. |
orchestrator_tlx_mental | VOX_ORCHESTRATOR_TLX_MENTAL | vox-orchestrator | active | 1.0 | NASA-TLX mental demand weight used in agent workload scoring. |
orchestrator_tlx_temporal | VOX_ORCHESTRATOR_TLX_TEMPORAL | vox-orchestrator | active | 1.0 | NASA-TLX temporal pressure weight used in agent workload scoring. |
orchestrator_tlx_trust_discount | VOX_ORCHESTRATOR_TLX_TRUST_DISCOUNT | vox-orchestrator | active | 0.0 | Trust-based discount applied to TLX workload score for high-reliability agents. |
orchestrator_toestub_gate | VOX_ORCHESTRATOR_TOESTUB_GATE | vox-orchestrator | active | false | Enable the toestub debug/inspection gate for orchestrator sessions. |
orchestrator_trust_auto_approve_min | VOX_ORCHESTRATOR_TRUST_AUTO_APPROVE_MIN | vox-orchestrator | active | 0.85 | Minimum trust score (0.0–1.0) required for auto-approve without human review. |
orchestrator_trust_ewma_alpha | VOX_ORCHESTRATOR_TRUST_EWMA_ALPHA | vox-orchestrator | active | 0.1 | EWMA smoothing factor for per-agent trust score updates. |
orchestrator_trust_gate_relax_enabled | VOX_ORCHESTRATOR_TRUST_GATE_RELAX_ENABLED | vox-orchestrator | active | false | Allow relaxing trust-gate checks for agents that exceed the reliability threshold. |
orchestrator_trust_gate_relax_min_reliability | VOX_ORCHESTRATOR_TRUST_GATE_RELAX_MIN_RELIABILITY | vox-orchestrator | active | 0.85 | Minimum reliability score (0.0–1.0) needed to be eligible for trust-gate relaxation. |
orchestrator_trust_provisional_threshold | VOX_ORCHESTRATOR_TRUST_PROVISIONAL_THRESHOLD | vox-orchestrator | active | 5 | Minimum completed tasks before an agent leaves provisional trust status. |
orchestrator_trust_trusted_threshold | VOX_ORCHESTRATOR_TRUST_TRUSTED_THRESHOLD | vox-orchestrator | active | 20 | Completed-task count at which an agent reaches fully-trusted status. |
orchestrator_urgent_rebalance_threshold | VOX_ORCHESTRATOR_URGENT_REBALANCE_THRESHOLD | vox-orchestrator | active | 3 | Queue depth at which an immediate (non-cooldown) rebalance is triggered. |
pre_push_refs | VOX_PRE_PUSH_REFS | vox-cli | active | — | Set by the lefthook pre-push hook so vox ci pre-push reads git’s pushed refs from stdin. |
rag_chunk_chars | VOX_RAG_CHUNK_CHARS | vox-search | declared | 4096 | Maximum character count per RAG markdown chunk. |
real_model_dir | VOX_REAL_MODEL_DIR | vox-populi | declared | — | Test-only. Quantized Qwen3 checkpoint directory for the ignored candle CPU real-checkpoint test. |
sandbox_cpus | — | vox-container-types | active | 2 | Default CPU quota for sandboxed container runs (RunOpts::sandboxed()). |
sandbox_memory | — | vox-container-types | active | 2g | Default memory cap for sandboxed container runs. |
sandbox_pids_limit | — | vox-container-types | active | 512 | Default process cap for sandboxed container runs. |
search_arxiv_url | VOX_SEARCH_ARXIV_URL | vox-search | active | — | Override base URL for the arXiv research lane API. |
search_candidate_depth | VOX_SEARCH_CANDIDATE_DEPTH | vox-search | active | 10 | Number of retrieval candidates per lane before fusion (minimum 1). |
search_deep_timeout_ms | VOX_SEARCH_DEEP_TIMEOUT_MS | vox-search | active | 4000 | Per-request timeout for the deep research lane, in milliseconds. |
search_default_lane | VOX_SEARCH_DEFAULT_LANE | vox-search | active | fast | Default research lane, “fast” or “deep”. |
search_enable_arxiv | VOX_SEARCH_ENABLE_ARXIV | vox-search | active | true | Enable the arXiv research source (1/true/yes). |
search_enable_openalex | VOX_SEARCH_ENABLE_OPENALEX | vox-search | active | true | Enable the OpenAlex research source (1/true/yes). |
search_enable_wikipedia | VOX_SEARCH_ENABLE_WIKIPEDIA | vox-search | active | true | Enable the Wikipedia research source (1/true/yes). |
search_fast_timeout_ms | VOX_SEARCH_FAST_TIMEOUT_MS | vox-search | active | 1500 | Per-request timeout for the fast research lane, in milliseconds. |
search_openalex_url | VOX_SEARCH_OPENALEX_URL | vox-search | active | — | Override base URL for the OpenAlex research lane API. |
search_tavily_url | VOX_SEARCH_TAVILY_URL | vox-search | active | — | Override base URL for the Tavily research API. |
search_wikipedia_url | VOX_SEARCH_WIKIPEDIA_URL | vox-search | active | — | Override base URL for the Wikipedia research lane API. |
secrets_vault_keyring_service | VOX_SECRETS_VAULT_KEYRING_SERVICE | vox-secrets | active | vox-secrets-vault | OS keychain service name holding the vault master key; tests override it to avoid the real entry. |
speech_test_force_backend_fail | VOX_TEST_FORCE_BACKEND_FAIL | vox-speech | declared | — | Test-only (cfg(test)). Forces ASR backend creation to fail to exercise the fallback path. |
wasm_skill_fuel | VOX_WASM_SKILL_FUEL | vox-plugin-runtime-wasm | active | 1000000000 | Fuel budget (instruction count) for WASM skill execution. |