turnstone/docs/settings.md

System Settings

See also: Settings Architecture diagram

The system settings feature provides database-backed configuration for server nodes. Settings are stored in the system_settings table and managed through the admin API or console Settings tab. This replaces config.toml for non-bootstrap settings on server entry points, while the CLI continues to read config.toml directly.

Overview

Settings follow a typed registry pattern: every storable setting has a SettingDef entry in settings_registry.py with type, default, description, validation constraints, and a restart_required flag. Unknown keys are rejected at the API boundary.

At runtime, ConfigStore loads all settings from storage into an in-memory cache. Reads are lock-free dict lookups on an immutable snapshot. Writes acquire a lock, persist to storage, and swap the cache atomically.

---

Precedence

Settings resolution differs between entry points:

Entry point	Chain
Server (turnstone-server)	CLI flag > ConfigStore > registry default
CLI (turnstone)	CLI flag > config.toml > argparse default

The server's apply_config() ignores config.toml sections that overlap with ConfigStore. A startup warning is logged for each overlapping key, directing users to the admin Settings API.

---

Per-Model Sampling Overrides

The global model.temperature, model.max_tokens, and model.reasoning_effort settings serve as cluster-wide defaults. Individual models can override these via per-model settings in the model_definitions table (admin Models tab).

Resolution order for sampling parameters:

Priority	Source
1 (highest)	Per-model override (set in Models tab)
2	Global default (set in Settings tab)
3	Registry default (code)

When a per-model override is NULL (empty in the UI), the global default is used. Switching models via /model <alias> re-resolves sampling parameters from the new model's overrides or global defaults.

Model backend authentication

Model definitions support three backend credential modes:

auth_mode	Identity sent to the model gateway
static	The definition's stored api_key.
entra_obo	A caller-delegated Entra access token minted from that user's captured OIDC credential.
entra_app	A shared app-identity token minted with Turnstone's OIDC client credentials.

Dynamic modes require an exact obo_audience resource App ID URI. Before an admin can save one, an operator must add that literal audience to model.auth_audience_allowlist (comma- or newline-separated). Wildcards and base-URL host matching are intentionally unsupported. Changing dynamic auth, its audience, or the gateway base_url also requires admin.mcp; service tokens do not bypass this capability-escalation gate.

entra_app is supported only with [oidc] obo_grant_profile = "entra". Judge, output-guard, perception, utility, and sub-agent lanes inherit the session's effective user for entra_obo. The perception memo is partitioned by that principal as well as alias and content hash, so a result authorized as one user cannot be served to another. Scheduled and wake-driven work retains the workstream owner even when no user is connected. Eval and optimizer lanes are registry-less development tools and therefore do not use dynamic model authentication.

entra_app is an explicit model-definition choice; Turnstone never changes a failed or ownerless entra_obo call into a client-credentials grant. An entra_obo call with no effective user always refuses. A dynamic alias without a real static key also always refuses instead of issuing its SDK-construction placeholder. When a real static key is explicitly configured, mint failures may use it by default; set model.auth_fail_closed = true to prohibit even that fallback. A refusal is not routed through the model fallback chain.

Dynamic token caches are encrypted in mcp_user_tokens, shared across nodes, and memoized on each host. Unlinking a user's OIDC identity purges their entra_obo rows and memo entries. entra_app rows belong to the shared __app__ identity and are not user-deprovisioned; after client-credential revocation, an already-minted app bearer remains usable until its recorded expiry.

obo_audience is literal and capped at 2048 characters. Environment-variable expansion is deliberately not applied, so the allow-list decision cannot vary by node or expand beyond the persisted boundary.

Responses output controls (per-model)

Models whose capability table declares Responses output controls expose two additional fields in the Models create/edit shelf:

Field	Stored capability	Values	Effect
Output verbosity	verbosity	low, medium, high	Controls answer length independently of reasoning effort.
Reasoning mode	reasoning_mode	standard, pro	Selects standard or higher-compute Pro execution without changing the model ID.

An empty selection means provider default and omits the capability key. Known GPT-5.6 models inherit support from the built-in table without persisting redundant support flags. An OpenAI-compatible model pinned to the Responses API can opt in with the supports_verbosity and supports_pro_mode capability tiles. Chat Completions and non-Responses providers do not surface or submit these controls.

Removed settings: model.name and model.context_window have been removed from ConfigStore. Model names and context windows are now configured per-model in the Models tab. A startup warning is logged if these keys appear in config.toml.

Reasoning persistence (per-model)

Two boolean flags on model_definitions (migration 052) control how reasoning text round-trips per model:

Flag	Default	Effect
surface_persisted_reasoning	True	Surface stored reasoning text on /history payloads so a page reload re-renders the reasoning bubble. Storage of reasoning bytes is independent of this flag — they ride in provider_data regardless.
replay_reasoning_to_model	False	Send stored reasoning blocks back to the provider on subsequent turns. Capability-gated: only takes effect when the model's ModelCapabilities.supports_reasoning_replay is also True. Set on canonical OpenAI gpt-5*/o-series and Anthropic Claude entries; unknown / local-server models default to False so an operator who flips the flag on a model whose API doesn't understand reasoning replay silently no-ops rather than 400-ing.

Edit both via the admin Models tab. See the architecture doc for the provider-side mechanics (Anthropic thinking, OpenAI Responses reasoning + include=["reasoning.encrypted_content"], synthetic reasoning_text for Chat Completions / vLLM / llama.cpp / Gemini-compat).

Task agent overrides

task_agent sub-sessions resolve independently from the conversation model so operators can pick a cheaper/faster model for autonomous loops:

Setting	Purpose
model.task_alias	Alias used for task_agent sub-sessions. Falls back to [model].agent_model in config.toml, then the session's active model.
model.task_effort	Reasoning effort for task_agent. Empty string means "inherit from the session".

Both are live-editable from the Settings tab and take effect on the next sub-agent invocation — no restart required.

---

Bootstrap vs ConfigStore

Bootstrap settings are required before storage is available (database connection, Redis, auth secrets, server bind address). These stay in config.toml and environment variables.

Category	Section	Where
API credentials	[api]	config.toml / env
Database	[database]	config.toml / env
Auth	[auth]	config.toml / env
Console bind	[console]	config.toml / env

ConfigStore settings are loaded from the database after storage initialization:

Section	Settings
model	default_alias, auth_audience_allowlist, auth_fail_closed, temperature, max_tokens, reasoning_effort, task_alias, task_effort
session	instructions, retention_days, compact_max_tokens, auto_compact_pct
tools	timeout, truncation, agent_max_turns, skip_permissions, search, search_threshold, search_max_results
server	workstream_idle_timeout, max_workstreams
cluster	node_fan_out_limit, mcp_max_servers
mcp	config_path, registry_url
ratelimit	enabled, requests_per_second, burst, trusted_proxies
health	backend_probe_interval, backend_probe_timeout, circuit_breaker_threshold, circuit_breaker_cooldown
judge	enabled, model, provider, base_url, api_key, confidence_threshold, max_context_ratio, timeout, read_only_tools, output_guard, redact_secrets, cancel_on_approval
interface	close_tab_action, theme
skills	discovery_url
memory	relevance_k, fetch_limit, max_content, nudge_cooldown, nudges

Settings are addressed by dotted key (e.g. memory.relevance_k). Each has a declared type (int, float, str, bool), optional min_value/max_value range, optional choices list, and an is_secret flag.

---

Storage

The system_settings table (migration 015) stores settings as JSON-encoded values with a composite primary key of (key, node_id):

Column	Type	Description
key	text	Dotted setting key (e.g. model.temperature)
value	text	JSON-encoded value
node_id	text	Node ID for per-node overrides (empty string = global)
is_secret	int	1 if the setting contains secrets
changed_by	text	Username of last editor
created	text	ISO timestamp
updated	text	ISO timestamp

Per-node overrides layer on top of global settings. When ConfigStore loads, it fetches global settings first, then overlays per-node values.

---

Admin API

Four endpoints on the console server, all requiring the admin.settings permission.

GET /v1/api/admin/settings

List all settings with their effective values, defaults, and metadata.

Response: 200

{
  "settings": [
    {
      "key": "model.temperature",
      "value": 0.7,
      "source": "storage",
      "type": "float",
      "description": "Sampling temperature",
      "section": "model",
      "is_secret": false,
      "node_id": "",
      "changed_by": "admin",
      "updated": "2026-03-14T10:00:00",
      "restart_required": false
    }
  ]
}

---

GET /v1/api/admin/settings/schema

Return the full registry catalog (all defined settings with metadata). Useful for building dynamic admin UIs.

Response: 200

{
  "schema": [
    {
      "key": "model.temperature",
      "type": "float",
      "default": 0.5,
      "description": "Sampling temperature",
      "section": "model",
      "is_secret": false,
      "min_value": 0.0,
      "max_value": 2.0,
      "choices": null,
      "restart_required": false
    }
  ]
}

---

PUT /v1/api/admin/settings/{key}

Update a setting. The value is validated against the registry (type coercion, range, choices). Secret settings (is_secret=true) cannot be written via the API -- they must be configured via config.toml or environment variables.

Path parameters:

Parameter	Type	Description
key	string	Dotted setting key (e.g. model.temperature)

Request body:

{
  "value": 0.7,
  "node_id": ""
}

Field	Type	Required	Default	Description
value	any	yes	--	New value (type-coerced against registry)
node_id	string	no	""	Node ID for per-node override

Response (success): 200

{
  "key": "model.temperature",
  "value": 0.7,
  "source": "storage",
  "type": "float",
  "description": "Sampling temperature",
  "section": "model",
  "is_secret": false,
  "node_id": "",
  "changed_by": "admin",
  "updated": "",
  "restart_required": false
}

Errors:

Status	Condition
400	Unknown key, invalid value, type mismatch, out of range
403	Secret setting (must use config.toml or env)

---

DELETE /v1/api/admin/settings/{key}

Reset a setting to its registry default by removing it from storage.

Path parameters:

Parameter	Type	Description
key	string	Dotted setting key

Query parameters:

Parameter	Type	Required	Default	Description
node_id	string	no	""	Node ID (empty = global)

Response (success): 200

{"status": "ok", "key": "model.temperature", "default": 0.5}

Response (not found): 404

{"error": "Setting 'model.temperature' has no stored value"}

---

Secret Settings

Settings with is_secret=True (currently only judge.api_key) are blocked from the write API with a 403 response. This prevents accidental exposure through the admin UI or audit logs. Secret settings must be configured via config.toml or environment variables.

The list endpoint masks secret values: stored secrets appear as "***" rather than their actual value.

---

Hot Reload

ConfigStore caches all settings in memory for fast, lock-free reads. To refresh the cache after external changes (e.g. direct database edits or cluster-wide propagation):

POST /v1/api/_internal/config-reload

This triggers ConfigStore.reload(), which re-reads all settings from storage and atomically swaps the cache. The version counter increments on every reload.

Behavior after reload:

• New workstreams pick up updated values immediately (via session_factory)
• Existing sessions keep their frozen configuration (settings are captured at workstream creation time, not read on every turn)
• Settings marked restart_required=True need a server restart to take effect

---

Migration from config.toml

On startup, warn_migrated_settings() scans config.toml for keys that are now managed by ConfigStore. Each overlap produces a warning:

WARNING config.toml [model] temperature is now managed via Settings API —
this value will be ignored. Use the admin Settings tab or
PUT /v1/api/admin/settings/model.temperature to configure.

To migrate:

1. Note the values from config.toml for sections that overlap with ConfigStore
2. Use PUT /v1/api/admin/settings/{key} or the console Settings tab to set each value
3. Remove the migrated sections from config.toml
4. Restart the server to verify no warnings

---

SDK

Python

from turnstone.sdk import TurnstoneConsole

with TurnstoneConsole("http://localhost:9090", token="tok_xxx") as admin:
    # List all settings with effective values
    result = admin.list_settings()
    for s in result["settings"]:
        print(f"{s['key']} = {s['value']} (source: {s['source']})")

    # Get the schema catalog
    schema = admin.get_settings_schema()

    # Update a setting
    admin.update_setting("model.temperature", value=0.7)

    # Update with per-node override
    admin.update_setting("model.temperature", value=0.3, node_id="node-2")

    # Reset to default
    admin.delete_setting("model.temperature")

TypeScript

import { TurnstoneConsole } from "@turnstone/sdk";

const admin = new TurnstoneConsole({
  baseUrl: "http://localhost:9090",
  token: "tok_xxx",
});

// List all settings
const result = await admin.listSettings();
for (const s of result.settings) {
  console.log(`${s.key} = ${s.value} (source: ${s.source})`);
}

// Get schema catalog
const schema = await admin.getSettingsSchema();

// Update a setting
await admin.updateSetting("model.temperature", { value: 0.7 });

// Reset to default
await admin.deleteSetting("model.temperature");

---

Architecture

See Settings Architecture diagram for the full data flow covering server startup, admin API writes, hot reload, and settings precedence.