Workspace Settings

Overview

A workspace setting is one administrative switch that changes how Logrite treats a whole workspace: how AI coding sessions are recorded and checked, what happens to a log line between a collector and the index, what a new collector collects, and what reaches the bell. There are 29 keys in five groups — coding.*, logging.*, ingest.*, collector.defaults.* and notify.*.

• The catalogue is code, not data. Every key, its type, default and bounds live in the static WorkspaceSettingCatalog (backend, service/settings/). The ingest shim, the AI coding policy, the collector binding and the Settings hub read keys by name, so a key is a contract: adding one is a code change, and keys are never renamed (a rename would silently reset every workspace that changed the old one).
• The table holds only what an admin changed. workspace_setting (migration mirror 2026-10-workspace-settings.sql) has one row per workspace per key: workspace_id, setting_key (VARCHAR(96)), value_json (the JSON of an already-validated value), updated_by, updated_at, UNIQUE (workspace_id, setting_key). No row = the catalogue default, so a new key never needs a data migration.
• Reads are cached for 30 s per workspace. A read is the catalogue defaults overlaid with the stored values. A write evicts the cache immediately and again after commit. Consumers that cache on top (the shim’s 15 s binding cache, the coding policy’s 60 s cache) add their own delay — see How long a change takes to apply.
• A stored value the current catalogue refuses (a range that was later tightened, or a key a later build dropped) reads back as the default and logs a warning; it is never handed to a consumer.

REST API

Method	Path	Who	Body / answer
GET	/api/v1/workspaces/{workspaceId}/settings	Any workspace member — session JWT or a Logrite API key (lgp_live_…)	{workspaceId, settings, catalogue}
PATCH	/api/v1/workspaces/{workspaceId}/settings	Admin or Owner	Body: a flat partial map {"<key>": value, …}. Answer: same shape as GET, with the new values

A non-member gets 404 workspace not found (or 403 where LOGRITE_CROSS_TENANT_NOT_FOUND=false). A member below Admin gets 403 changing workspace settings requires workspace admin — the role is checked before any key is looked at, so an outsider cannot probe which keys exist from the 400s.

Response. settings holds every key (stored value, else default). catalogue describes every key as {key, type, default, min?, max?, values?, maxItems?, maxLength?}; type is one of bool, int, enum, string, list. Read the catalogue before writing — it is what the Settings hub uses for its bounds.

curl -s "https://leapfrogapi.deployd.io/api/v1/workspaces/$WORKSPACE_ID/settings" \
  -H "Authorization: Bearer $LOGRITE_API_KEY"
# {"workspaceId":3103,
#  "settings":{"coding.enabled":true,"coding.capture":"findings",…,"notify.coding.bell":true},
#  "catalogue":[{"key":"coding.enabled","type":"bool","default":true},
#               {"key":"coding.capture","type":"enum","default":"findings","values":["metadata","findings","extended"]},
#               {"key":"coding.retentionDays","type":"int","default":30,"min":7,"max":90}, …]}

curl -s -X PATCH "https://leapfrogapi.deployd.io/api/v1/workspaces/$WORKSPACE_ID/settings" \
  -H "Authorization: Bearer $LOGRITE_API_KEY" -H "Content-Type: application/json" \
  -d '{"ingest.sampling.enabled": true, "ingest.sampling.percent": 25, "coding.capture": null}'

PATCH semantics.

• All or nothing. Every key in the body is validated before any is written; one bad key refuses the whole request and nothing is stored.
• null resets a key to its default (the stored row is deleted). In the example above coding.capture goes back to findings.
• Values are normalised before storage: strings and list items are trimmed; enum values and list items from a fixed set match case-insensitively and are stored in the catalogue’s own spelling ("error" → "ERROR"); list items are de-duplicated; a whole number sent as 25.0 is accepted, a fraction is not.
• An empty body is a no-op that answers the current values.
• Each successful PATCH writes one governance-audit row, action settings.update, with the actor’s email and {"keys": […]} — the keys changed, not their values.

Status	Body	When
400	{"error":"unknown_setting","key":"coding.foo"}	The key is not in the catalogue
400	{"error":"invalid_setting","key":"ingest.sampling.percent","rule":"must be a whole number between 1 and 100"}	The value breaks the key’s rule
403	—	Caller is a member but not Admin/Owner (PATCH only)
404	—	Caller is not a member of the workspace

The rule strings the backend answers with:

• must be true or false
• must be a whole number between <min> and <max>
• must be one of: … / every item must be one of: …
• must be a string / must be a list of strings / every item must be a non-empty string
• at most <n> items
• must be at most 64 characters / every item must be at most 64 characters
• must be a field name: a letter or _ followed by letters, digits, _ . or - (also with the every item prefix)

Field names are used inside regexes downstream

logging.correlationField, logging.piiFields and ingest.redaction.fields must match [A-Za-z_][A-Za-z0-9_.-]* and be at most 64 characters; the two lists hold at most 50 items. An empty logging.correlationField means “not set” and is allowed.

Every key

The “Read by” column names the code that acts on the key. Defaults are what a workspace that never changed the key gets.

coding.* — AI coding sessions

Read by the Claude Code plugin through the AI coding policy (GET {mcpUrl}/coding/policy, which carries the coding.* and logging.* values; the plugin caches it and uses built-in copies of these defaults until a policy has been fetched), and by the backend where noted.

Key	Type	Default	Allowed	What it does / read by
coding.enabled	bool	true	—	Master switch for recording. MCP server coding ingest: when off, POST /coding/events answers 202 {"accepted":0,"disabled":true} and stores nothing. Plugin: does not send turn events or commit-scan events when off; it still checks code locally
coding.capture	enum	findings	metadata, findings, extended	What is recorded per finding. MCP server coding ingest enforces it whatever the plugin sends: metadata drops each finding’s message, snippet and path (a pathHash is kept). extended currently records the same as findings (300-character messages, 200-character masked snippets). Shown on the AI Coding issue detail as captureLevel
coding.retentionDays	int	30	7–90	How long AI coding session data is kept. Backend CodingRetentionJob (cron 0 30 3 * * *) deletes coding-events-<ws> events and stored evidence (masked snippets, paths) older than this; issue rows and their counts, mute and fixed state are kept
coding.heal	bool	true	—	Plugin Stop hook: hand the agent the issues in the files it changed and ask it to fix them before it stops. Never blocks
coding.healRounds	int	1	0–2	Plugin: how many times one turn can be sent back (0 = never). Never while the Stop hook is already active
coding.healMinSeverity	enum	medium	low, medium, high, critical	Plugin: issues below this are recorded but not sent back to the agent
coding.precommit	bool	true	—	Plugin pre-commit hook (installed by /logrite:init only when the developer agrees): when off, the hook does not check. It only warns — the commit always goes through
coding.scan.compliance	bool	true	—	Backend coding policy includes the workspace’s compliance-framework rules only when on; the plugin applies them to lines the agent adds
coding.scan.logging	bool	true	—	Policy includes the logging-standard rules (credential or PII in a log call, missing correlation field) only when on
coding.scan.business	bool	true	—	Policy includes the workspace’s business rules on the AI coding plane only when on
coding.notifyMinSeverity	enum	high	low, medium, high, critical	Backend CodingIssueNotifier: a new or returning AI coding issue below this severity is announced neither on Discord nor on the bell

logging.* — logging standards for AI-written code

Key	Type	Default	Allowed	What it does / read by
logging.correlationField	string	"" (not required)	field name, ≤ 64 chars	Backend coding policy builds the “log statement omits the required correlation field” rule from it (e.g. traceId). Empty = no such rule
logging.piiFields	list	email, ssn, password, phone, dob, cardNumber	field names, ≤ 50 items, each ≤ 64 chars	Policy’s “log statement names a person” rule; the plugin also uses the list to mask those fields’ values in finding snippets. When an effective AI-coding policy of the workspace includes HIPAA, the policy appends fullName, full_name, birthDate, birth_date, dateOfBirth, date_of_birth, mrn (2026-10-09; the stored value is unchanged)

ingest.* — ingest-time redaction and sampling

Read by the ingest shim (POST /ingest) on every line from a collector agent or the ingest endpoint, via the agent resolve binding’s ingestPolicy. Not applied to OTLP/Data Prepper or the Bulk API. Full behaviour: Ingest-time redaction and sampling.

Key	Type	Default	Allowed	What it does
ingest.redaction.enabled	bool	false	—	Mask matching values before the line is stored; a changed line carries redacted: true
ingest.redaction.kinds	list	every kind except longDigits	urlCredential, credentialPair, email, ssn, mrn, card, iban, phone, dob, secretKey, opaqueToken, longDigits (≤ 12 items)	Which value patterns are masked. longDigits (9+ digit numbers) is off by default because it also masks order and trace IDs. mrn, phone and dob were added 2026-10-09 (patterns in Relay compliance → Redaction kinds)
ingest.redaction.fields	list	password, accessToken, refreshToken, otp, pin, last4, ssn, ccNumber, ccNbr, creditCardNumber, creditCardNbr	field names, ≤ 50 items, each ≤ 64 chars	Field names whose following value is masked (password=hunter2 → password=***)
ingest.sampling.enabled	bool	false	—	Store only a share of lines; kept lines carry sampleRate
ingest.sampling.percent	int	100	1–100	Share of lines kept at levels not in alwaysKeep
ingest.sampling.alwaysKeep	list	FATAL, ERROR, WARN	FATAL, ERROR, WARN, INFO, DEBUG, TRACE (≤ 6 items)	Levels never sampled out

warden.* and cortex.* — privacy switches (2026-10-09)

Admin-written. Every change to one of these six keys writes the settings.update GovernanceAudit row with details.privacyChanges naming each key with its old and new value, so turning masking off can be found later. Behaviour: AI Warden → Redaction switches and AI service → Cortex masking.

Key	Type	Default	Allowed	What it does / read by
warden.redaction.enabled	bool	true	—	Relay: false stores and shows Warden text as sent — previews, raw body, tool-call timeline, block message, compliance snippets, judge reasons. Detection and verdicts are unaffected
warden.redaction.kinds	list	every kind except longDigits	same kinds as ingest.redaction.kinds	Kinds the relay masks. The raw body masks only the credential kinds (urlCredential, credentialPair, secretKey, opaqueToken) that are listed
warden.redaction.fields	list	same as ingest.redaction.fields	field names, ≤ 50 items	Field names whose value the relay masks
warden.sensitiveRawCapture	bool	false	—	“Keep the raw text of requests blocked for personal or health data. Turn on only when your model runs inside your own network.” When false, rows blocked or flagged for PHI/PII keep no raw text
cortex.redaction	enum	auto	auto, on, off	Cortex: mask identifiers in answers. auto masks when HIPAA is enabled in the workspace (a policy that is not Off holds it, or a scan profile uses it)
cortex.redaction.kinds	list	all	email, ssn, mrn, card, iban, phone, dob	Which identifiers Cortex masks when masking is on

The relay reads the first four through the relay-key binding (redaction {enabled, kinds, fields, patterns} and sensitiveRawCapture); a binding without the block (older backend or cached binding) keeps the previous fixed masks.

collector.defaults.* — what a new collector collects

Read by the backend when a collector is created (POST /api/v1/agents): any capability or interval the request leaves out (null) takes these values, and the New Collector Agent modal pre-fills its fields from them (fields you already edited are left alone). Existing collectors are not changed by editing a default — use Apply to existing collectors (POST /api/v1/agents/apply-defaults, see Collector agents → Live config at check-in).

Key	Type	Default	Allowed	Capability
collector.defaults.intervalMinutes	int	15	1–1440	Check-in / ship interval (minutes)
collector.defaults.heartbeat	bool	true	—	Heartbeat each cycle
collector.defaults.systemDetails	bool	true	—	System details (system_metrics host doc)
collector.defaults.journald	bool	false	—	systemd journal
collector.defaults.varlog	bool	false	—	/var/log files
collector.defaults.docker	bool	false	—	Docker containers
collector.defaults.k8s	bool	false	—	Kubernetes pods
collector.defaults.syslog	bool	false	—	Syslog receiver

Heartbeat and System details now default to on for API-created collectors

Before this release a POST /api/v1/agents that omitted heartbeatEnabled or collectSystemDetails created the collector with them off. They now take collector.defaults.heartbeat / collector.defaults.systemDetails, which default to on. Send false explicitly if you want them off.

notify.* — the bell

Key	Type	Default	What it does / read by
notify.warden.bell	bool	true	Frontend notification sync: when off, AI Warden blocked/flagged requests are not added to the bell. Evaluated in each browser; an unreadable store counts as on
notify.coding.bell	bool	true	Backend CodingIssueNotifier: when off, new or returning AI coding issues are not posted to the bell (Discord is unaffected; coding.notifyMinSeverity gates both)

gatekeeper.* — the pull-request gate (2026-10-15)

Key	Type	Default	Allowed	What it does / read by
gatekeeper.coverageThreshold	int	80	0–100	Backend GatekeeperThreshold: the log-coverage threshold a Gatekeeper check is measured against when the request sends none. Precedence: the request’s logCoverageThreshold → a stored workspace row (the workspace of the latest scan’s project; thresholdSource: "workspace") → the server property. Edited (Admin+) in Rules & Policies → Where rules apply → Code scans & log generation (“Pull-request gate”)

How long a change takes to apply

Consumer	Delay after a PATCH
Settings hub, workspace_settings get	Immediate (the writer’s cache is evicted)
Ingest shim (ingest.*)	Up to the shim’s agent-binding cache, 15 s (SHIM_TOKEN_TTL_MS), plus up to 30 s on a backend instance that did not serve the PATCH
Collectors (collector.defaults.*)	Only new collectors, or existing ones after Apply to existing collectors — each picks it up at its next check-in
Claude Code plugin (coding.*, logging.*)	From the next session; the policy endpoint caches 60 s per workspace

The 30 s settings cache is per backend instance; only the instance that served the PATCH evicts its copy.

The Settings hub

Four tabs in Settings read and write these keys (/settings/aiCoding, /settings/logIngestion, /settings/privacy, /settings/notifications). Every member can open them and see what is in effect; only an Admin or Owner can change a switch — others see “Only a workspace admin can change these settings. You can see what is in effect.” Every switch is the shared LogriteSwitch control, and each change is one PATCH.

AI coding (aiCoding)

Card	Anchor	Keys
Recording	#setting-ai-coding-recording	coding.enabled (“Record AI coding sessions”), coding.capture (three side-by-side cards: Metadata only / Masked findings / Extended, each listing what is and is not recorded), coding.retentionDays (“Keep sessions for”)
Fixing in-session	#setting-ai-coding-heal	coding.heal, coding.healRounds (Never / Once / Twice), coding.healMinSeverity
Commit check	#setting-ai-coding-precommit	coding.precommit (“Warn at commit”)
What to check	#setting-ai-coding-checks	coding.scan.compliance, coding.scan.logging, coding.scan.business
Logging standards	#setting-ai-coding-logging	logging.correlationField, logging.piiFields
Privacy	#setting-ai-coding-privacy	Read-only: what is never recorded, and the per-repository opt-out ("recordSessions": false)
Recording tokens	#setting-ai-coding-tokens	Not a setting. Lists the workspace’s write-only lgsi_live_ tokens that /logrite:init mints — one per workspace, shared by every repository on a machine (2026-10-09; earlier plugin versions minted one per repository) — (label, masked token lgsi_live_… + last four, created by, created, last used, Active/Revoked) and revokes them; a workspace admin or the token’s creator may revoke. The card says a revoked token “stops working at once”; in practice the MCP server caches a valid token for up to 60 s, so the plugin stops at its first flush after that. Recorded sessions are kept — see AI Coding → the session-ingest token
Setup	#setting-ai-coding-setup	Read-only: the plugin install one-liner and /logrite:init

Log ingestion (logIngestion)

Card	Anchor	Keys
Redaction	#setting-ingest-redaction	ingest.redaction.enabled (“Redact before storing”), one switch per kind (ingest.redaction.kinds, with an example of what each kind masks), ingest.redaction.fields. Footnote: redaction and sampling apply to collectors and the ingest endpoint, not to OTLP exports
Sampling	#setting-ingest-sampling	ingest.sampling.enabled, ingest.sampling.percent (“Keep N% of lines at the other levels”), one switch per level (ingest.sampling.alwaysKeep)
Retention	#setting-ingest-retention	Not a workspace setting: the M-26-14 retention policy (hot/warm/cold tiers, searchable/retrievable windows), the same form as the M-26-14 page, saved through PUT /api/v1/retention/policy. Maintainer or above may save it
Collector defaults	#setting-ingest-collector-defaults	collector.defaults.intervalMinutes (“Check-in interval”) and one switch per capability, plus the Apply to existing collectors button (confirm dialog “Apply these defaults to every collector?”)

Privacy (privacy, 2026-10-09)

Card	Anchor	Keys
AI Warden redaction	#setting-privacy-warden-redaction	warden.redaction.enabled (“Mask what AI Warden stores”), one switch per kind (warden.redaction.kinds, now including MRN, phone and date of birth), warden.redaction.fields. Applies to requests captured after the change
Raw text of requests blocked for personal data	#setting-privacy-sensitive-raw	warden.sensitiveRawCapture (“Keep the raw text”)
Cortex answers	#setting-privacy-cortex	cortex.redaction (row “Mask identifiers”, Auto / On / Off — “Auto masks identifiers in workspaces whose policies use HIPAA.”), cortex.redaction.kinds. A line points to Log ingestion → Redaction for stored logs

Notifications (notifications)

Card	Anchor	What
In-app notifications	#setting-notifications-in-app	notify.warden.bell (“AI Warden”), notify.coding.bell (“AI coding”), coding.notifyMinSeverity (“AI coding: minimum severity — for the bell and Discord alike”)
Discord	#setting-notifications-discord	The Discord channel card, moved here from Settings → Integrations. Not a workspace setting; Maintainer or above may manage it
Email preferences	#setting-email-preferences	Moved here from Settings → General. Per-user, not per-workspace (saved through the user-settings API): Log analysis report, Critical issues alert, Code generation success, Welcome email. Security emails (verification, password reset, sign-in codes, new-device alerts) are always sent

Settings → Connections (MCP / OAuth connections) lost its Workspace column (it always showed the current workspace) and merged Created and Last used into one Activity column (“Created …” over “Last used …” / “Never used”).

All of these cards are reachable from the settings search box.

MCP tool: workspace_settings

The MCP server exposes the store as one grouped tool in the Workspace family. The workspace is always the one the credential belongs to — there is no workspace argument, so a write cannot be aimed at a workspace the model merely knows the number of.

Action	Scope	Arguments	Backend call
get	workspace:read	none	GET /api/v1/workspaces/{ws}/settings — every value plus the catalogue
set	settings:manage	settings: a flat map, e.g. {"coding.capture": "metadata", "ingest.sampling.percent": 25}	PATCH the same path; still needs the Admin or Owner role on the backend

set with a missing or empty settings map is refused before any call. An unknown key or a bad value refuses the whole change and names the key (the backend’s 400).

The OAuth consent text for settings:manage (“Change workspace settings”) now reads “Change workspace settings: AI coding, log ingestion and notifications.” (it previously described add-ons, integrations, observability connections and retention).