Generated config reference

This page is generated from the same config field registries that power mono-agent config, recipe output, and the JSON Schema. Do not edit this table by hand; run pnpm run generate:config-reference.

Schema: https://raw.githubusercontent.com/robertsreberski/mono-agent/main/packages/app/schema/mono-agent.config.schema.json

JSON key	Type	Env override	Default	Example	Notes
agent.name	string	--	unset	Research Partner	Public display identity used for trace labels and default A2A metadata; never used in paths or service ids.
artifacts.dir	string	--	.mono-agent/artifacts	.mono-agent/artifacts	Configures dir for the artifacts section.
artifacts.memoryRetention.dryRun	boolean	--	false	true	Configures memoryRetention.dryRun for the artifacts section.
artifacts.memoryRetention.maxAgeDays	integer	--	7	7	Configures memoryRetention.maxAgeDays for the artifacts section.
artifacts.memoryRetention.maxCount	integer	--	5000	5000	Configures memoryRetention.maxCount for the artifacts section.
artifacts.replyFiles.maxFileBytes	integer	--	20971520	20971520	Maximum bytes per published reply file (default and maximum 20 MiB); larger values cannot be delivered safely by all channels.
artifacts.replyFiles.maxStorageBytes	`integer	unlimited`	--	2147483648	unlimited
artifacts.retention.dryRun	boolean	--	false	true	Configures retention.dryRun for the artifacts section.
artifacts.retention.maxAgeDays	integer	--	365	365	Configures retention.maxAgeDays for the artifacts section.
artifacts.retention.maxCount	integer	--	50000	50000	Configures retention.maxCount for the artifacts section.
channels.plugins	array	--	[]	[{"package":"@mono-agent/channel-whatsapp","config":{"enabled":true}}]	External channel plugin envelopes loaded by package name.
concurrency.maxConcurrentRuns	integer	--	unset	1	Configures maxConcurrentRuns for the concurrency section.
concurrency.maxPendingRuns	integer	--	unset	1	Configures maxPendingRuns for the concurrency section.
context.identityPath	string	--	required	./IDENTITY.md	Configures identityPath for the context section.
context.selectedSkills	string[]	--	[]	["example"]	Configures selectedSkills for the context section.
context.skillDisclosure	string	--	full	full	Configures skillDisclosure for the context section.
context.skillMaxBytes	integer	--	48000	48000	Configures skillMaxBytes for the context section.
context.skillsRoot	string	--	unset	example	Configures skillsRoot for the context section.
context.soulPath	string	--	unset	example	Configures soulPath for the context section.
continuations.detachedServices	array	--	[]	[{"name":"work-control","tokenEnv":"WORK_CONTROL_CONTINUATION_TOKEN"}]	Detached service names and the environment variable holding each bearer; raw tokens never belong in config.
continuations.enabled	boolean	--	true	true	Enables the host-owned durable continuation service when the block is configured.
continuations.host	string	--	127.0.0.1	127.0.0.1	Loopback bind host; non-loopback values are rejected.
continuations.limits.deliveryTimeoutMs	integer	--	120000	120000	Hard native-delivery and history-only commit timeout; ambiguous sends are never replayed automatically.
continuations.limits.maxActivePerOrigin	integer	--	500	500	Admission ceiling for one immutable run or detached-route claim origin.
continuations.limits.maxActiveRecords	integer	--	10000	10000	Global admission ceiling for non-terminal durable continuations.
continuations.limits.maxConcurrent	integer	--	16	16	Maximum independently tracked continuation workers; one hung provider cannot occupy the whole service.
continuations.limits.operatorPageSize	integer	--	100	100	Maximum keyset-paginated records returned by one operator list request.
continuations.limits.synthesisTimeoutMs	integer	--	600000	600000	Hard synthesis timeout; an ambiguous timeout is dead-lettered and never synthesized twice.
continuations.namedRoutes	object	--	{}	{"verification":{"mode":"capture","conversationId":"slack:D123"}}	Host-owned detached delivery policies: notify_if_actionable, capture, or silent.
continuations.port	integer	--	4319	4319	Fixed loopback continuation service port (1-65535); persisted result/status URLs and the operator CLI remain valid across restarts.
continuations.retention.capturedTextMaxAgeMs	integer	--	2592000000	2592000000	Maximum age in milliseconds for retained captured synthesis text.
continuations.retention.capturedTextMaxRecords	integer	--	1000	1000	Maximum delivered capture continuations whose synthesized text remains retrievable.
continuations.retention.terminalMaxAgeMs	integer	--	31536000000	31536000000	Maximum age in milliseconds for terminal continuation tombstones.
continuations.retention.terminalMaxRecords	integer	--	50000	50000	Maximum retained terminal metadata/idempotency tombstones after payload compaction.
continuations.stateDir	string	--	.mono-agent/continuations	.mono-agent/continuations	Owner-only per-record continuation store and token-derivation secret.
cron.conversationId	string	MONO_AGENT_CRON_CONVERSATION_ID	unset	example	Configures conversationId for the cron section.
cron.dir	string	MONO_AGENT_CRON_DIR	cron	cron	Configures dir for the cron section.
cron.effort	string	MONO_AGENT_CRON_EFFORT	unset	example	Configures effort for the cron section.
cron.enabled	boolean	MONO_AGENT_CRON_ENABLED	false	true	Enables the cron capability.
cron.expression	string	MONO_AGENT_CRON_EXPRESSION	unset	example	Configures expression for the cron section.
cron.jobs	array	MONO_AGENT_CRON_JOBS_JSON	[]	[{"id":"daily","expression":"0 8 * * *","prompt":"Summarize the overnight queue."}]	Inline scheduled jobs. Folder-based cron jobs still merge from cron.dir.
cron.model	string	MONO_AGENT_CRON_MODEL	unset	example	Configures model for the cron section.
cron.notify	boolean	MONO_AGENT_CRON_NOTIFY	false	false	Configures notify for the cron section.
cron.notifyConversationId	string	MONO_AGENT_CRON_NOTIFY_CONVERSATION_ID	unset	example	Configures notifyConversationId for the cron section.
cron.notifyFailureCooldownHours	integer	MONO_AGENT_CRON_NOTIFY_FAILURE_COOLDOWN_HOURS	6	6	Configures notifyFailureCooldownHours for the cron section.
cron.operatorActions.enabled	boolean	MONO_AGENT_CRON_OPERATOR_ACTIONS_ENABLED	false	true	Allows API-key-authenticated, explicitly confirmed run-now and runtime enable/disable actions. Defaults off and never rewrites cron config sources.
cron.preflight	string[]	MONO_AGENT_CRON_PREFLIGHT_JSON	unset	["node","scripts/check-queue.mjs"]	Explicit argv evaluated before the model responder. {"run":false} ends the firing as skipped_gate with no model turn; {"run":true,"input":"…"} runs the job with the input appended to its prompt. Any failure (non-zero exit, signal, timeout, malformed verdict) fails open with the plain prompt. Never a shell line.
cron.preflightTimeoutMs	integer	MONO_AGENT_CRON_PREFLIGHT_TIMEOUT_MS	5000	5000	Wall-clock bound for one preflight evaluation before it is killed and fails open. Positive integer milliseconds, default 5000, capped at 60000; separate from maxRunMs.
cron.prompt	string	MONO_AGENT_CRON_PROMPT	unset	example	Configures prompt for the cron section.
cron.timezone	string	MONO_AGENT_CRON_TIMEZONE	UTC	UTC	Configures timezone for the cron section.
interaction.askUser.timeoutMs	integer	--	600000	600000	Maximum wait for one AskUser interaction (one to five questions); set null to disable automatic expiry.
interaction.bridge.host	string	--	127.0.0.1	127.0.0.1	Bind host for the app-owned AskUser/tool-progress bridge. Defaults to loopback; keep it local because non-loopback values are not rejected.
interaction.bridge.port	integer	--	0	0	Bridge port. 0 chooses an ephemeral port.
interaction.progress.enabled	boolean	--	true	true	Whether tool progress posts are relayed to channel status messages.
memory.backend	string	--	bujo	bujo	Configures backend for the memory section.
memory.capture.cron	boolean	--	unset	false	Set false to skip automatic capture and raw audit for host-identified cron turns (including run-now); other turns and explicit Remember writes are unchanged. When Remember is enabled, this turn gets host-injected manual memory guidance. Unset preserves capture.
memory.capture.focus	string	--	unset	Keep durable coding preferences and verified lessons; skip PR and CI status.	Operator guidance (at most 2048 UTF-8 bytes) narrows BuJo capture extraction and review; it cannot override host safety or the strict JSON contract. Requires mode bujo and writeMode capture.
memory.capture.intentLifecycle	boolean	--	false	true	BuJo capture-only, default-off owner-supported intention state and inclusive civil end proposals. Reuses note status and due=; state changes supersede, never infer completion from a date. Tasks stay authoritative in the task app.
memory.capture.only	string[]	--	unset	["preference","lesson"]	Keep automatic capture memories only when a host-accepted label matches one of these kinds. An empty array drops all automatic captures; unset preserves current behavior. Remember writes are unaffected. Requires mode bujo and writeMode capture.
memory.capture.reconcileModel	string	--	unset	openai-codex:gpt-6-sol	Optional validated agent-host runtime model reference for the capture reconciliation classifier only. Unset uses memory.llm for extraction, review and reconciliation. Requires mode bujo and writeMode capture; failures do not fall back to the capture model.
memory.capture.webhook	boolean	--	unset	false	Set false to skip automatic capture and raw audit for host-identified webhook turns; other turns and explicit Remember writes are unchanged. When Remember is enabled, this turn gets host-injected manual memory guidance. Unset preserves capture.
memory.consolidation.cron	string	--	0 */2 * * *	0 */2 * * *	Configures consolidation.cron for the memory section.
memory.consolidation.enabled	boolean	--	true	true	Enables the memory capability.
memory.embeddings.apiKey	string	--	unset	example	Secret value for memory.embeddings.apiKey; prefer the env override.
memory.embeddings.apiKeyEnv	string	--	unset	example	Environment-variable name containing an optional provider bearer token; an explicitly declared name must resolve before memory starts.
memory.embeddings.circuitBreaker.cooldownMs	integer	--	30000	30000	Configures embeddings.circuitBreaker.cooldownMs for the memory section.
memory.embeddings.circuitBreaker.failureThreshold	integer	--	3	3	Configures embeddings.circuitBreaker.failureThreshold for the memory section.
memory.embeddings.dim	integer	--	unset	1	Configures embeddings.dim for the memory section.
memory.embeddings.endpoint	string	--	unset	example	Provider service root. LM Studio uses /v1/embeddings and defaults to http://localhost:1234.
memory.embeddings.instructions	string	--	auto	auto	Query/document instruction preset: auto (per model when an index is built; an existing index keeps its prefixes), search, none, query, or qwen3. Changing it requires the stopped-agent memory rebuild.
memory.embeddings.model	string	--	unset	nomic-embed-text:v1.5	Configures embeddings.model for the memory section.
memory.embeddings.provider	string	--	unset	ollama	Embedding service used by Journal/BuJo memory: ollama, lmstudio, or openai.
memory.embeddings.timeoutMs	integer	--	10000	10000	Configures embeddings.timeoutMs for the memory section.
memory.llm.endpoint	string	--	unset	example	Configures llm.endpoint for the memory section.
memory.llm.model	string	--	unset	openai-codex:gpt-5.6-terra	Configures llm.model for the memory section.
memory.llm.provider	string	--	unset	agent-host	Configures llm.provider for the memory section.
memory.llm.timeoutMs	integer	--	60000	60000	Configures llm.timeoutMs for the memory section.
memory.llm.trace	boolean	--	true	true	Configures llm.trace for the memory section.
memory.maxBytes	integer	--	64000	64000	Configures maxBytes for the memory section.
memory.mode	string	--	lite	journal	Configures mode for the memory section.
memory.path	string	--	unset	./.mono-agent/memory	Configures path for the memory section.
memory.profile.enabled	boolean	--	false	true	BuJo-only, default-off deterministic owner-stated label profile, at most 600 Unicode code points. Independent of embeddings. Warm provider context suppresses unchanged profile/served lines using invocation receipts; receipt loss conservatively suppresses automatic reinjection until a cold reseed.
memory.recall.intentExpiry	boolean	--	false	true	BuJo-only, default-off: exclude all dated notes and done/dropped notes from automatic sections, including unexpired/ambiguous legacy dates. Deliberate recall interprets reviewed note due= as an inclusive civil end. Requires reviewed curate audit before enablement; independent of intentLifecycle; no backfill.
memory.recall.recency	boolean	--	false	true	BuJo-only, default-off bounded secondary recency ranking for relevance-qualified deliberate event/unlabelled-note hits: at most 0.02, 30-day half-life. Facts/preferences never decay; base scores, floors and automatic retrieval are unchanged.
memory.recallTool.enabled	boolean	--	true	true	Enables explicit memory reads: targeted MemoryRecall for every backend and chronological MemoryJournal when a local store supports it. Automatic recall is unchanged.
memory.rememberTool.enabled	boolean	--	true	true	Enables the memory capability.
memory.writeMode	string	--	disabled	disabled	Configures writeMode for the memory section.
monitors	object	--	unset	{}	Deprecated compatibility object. The whole object, including unknown nested keys, is accepted and ignored. Use background process jobs for finite work.
openaiApi.allowNonLoopback	boolean	MONO_AGENT_OPENAI_API_ALLOW_NON_LOOPBACK	false	true	Configures allowNonLoopback for the openaiApi section.
openaiApi.apiKey	string	MONO_AGENT_OPENAI_API_KEY	unset	env:MONO_AGENT_OPENAI_API_KEY	Secret value for openaiApi.apiKey; prefer the env override.
openaiApi.basePath	string	MONO_AGENT_OPENAI_API_BASE_PATH	/v1	/v1	Configures basePath for the openaiApi section.
openaiApi.enabled	boolean	MONO_AGENT_OPENAI_API_ENABLED	false	true	Enables the openaiApi capability.
openaiApi.host	string	MONO_AGENT_OPENAI_API_HOST	127.0.0.1	127.0.0.1	Configures host for the openaiApi section.
openaiApi.modelId	string	MONO_AGENT_OPENAI_API_MODEL_ID	agent	agent	Configures modelId for the openaiApi section.
openaiApi.port	integer	MONO_AGENT_OPENAI_API_PORT	0	0	Configures port for the openaiApi section.
peers	object	--	none	{"finance":{"sourceId":"finance-ai"}}	Named local ACP peers for PeerAgent. Each sourceId must resolve to a running compatible mono-agent. Names and source IDs must be unique; this does not grant the peer owner approval.
processJobs.enabled	boolean	--	false	true	Opt in to owner-private Pi-native Exec/Bash background process jobs (unsupported on Windows).
processJobs.maxActivePerConversation	integer	--	2	2	Maximum simultaneously running or starting jobs per conversation (compiled cap 8); waiting jobs use the global queue. Nested child launches may borrow their running ancestor’s conversation slot.
processJobs.maxChainDepth	integer	--	4	4	Maximum host-owned background wake chain depth (compiled cap 64).
processJobs.maxConcurrent	integer	--	4	4	Maximum simultaneously running process jobs (compiled cap 32).
processJobs.maxOutputBytes	integer	--	1048576	1048576	Combined retained process-output ceiling (compiled cap 8 MiB).
processJobs.maxQueueAgeMs	integer	--	300000	300000	Maximum admission-to-spawn queue age (compiled cap one hour).
processJobs.maxQueued	integer	--	8	8	Maximum waiting process jobs globally, including jobs waiting on a conversation slot (compiled cap 64).
processJobs.maxRuntimeMs	integer	--	1800000	1800000	Host runtime ceiling per owned process group (compiled cap 24 hours; calls may narrow it).
processJobs.previewChars	integer	--	2000	2000	Bound for redacted wake/operator output previews (compiled cap 8000; calls may narrow it).
processJobs.retention.artifactMaxBytes	integer	--	268435456	268435456	Aggregate retained terminal output artifact budget (compiled cap 1 GiB).
processJobs.retention.maxAgeMs	integer	--	604800000	604800000	Maximum terminal process-job record age (compiled cap 30 days).
processJobs.retention.maxRecords	integer	--	1000	1000	Maximum retained terminal process-job records (compiled cap 10000).
processJobs.stateDir	string	--	.mono-agent/process-jobs	.mono-agent/process-jobs	Agent-root-confined owner-private process-job records and artifacts; must be disjoint from every restart —clear-sessions purge root.
processJobs.unsafeAllowUnprotectedState	boolean	--	false	false	Dangerous trusted-host opt-in: with explicit sandbox.mode=off and Pi-only routes, retain ProcessJobs state without synthetic SRT protection; state and the operator secret become model-accessible.
providers	object	--	unset	{"openai-codex":{},"ollama":{"baseUrl":"http://localhost:11434"}}	Provider-id map that widens or narrows the selectable Pi model catalog.
providers.piAuthPath	string	--	unset	~/.pi/agent/auth.json	Configures piAuthPath for the providers section.
providers.piNative.cacheRetention	string	--	long	long	Anthropic Messages cache retention (short or long; default long). Short opts out. JSON > long; the resolved value overrides ambient PI_CACHE_RETENTION. Long requires model support: 1h writes cost 2× normal input, reads 0.1×; short writes cost 1.25×. No guaranteed hit.
providers.piNative.maxRetryDelayMs	integer	--	60000	60000	Ceiling for provider-requested retry delays in milliseconds (default 60000). Nonpositive embedder values use the default; longer requested delays fail.
providers.piNative.piMaxRetries	integer	--	2	2	Provider SDK-policy retries before a response starts (0–8, default 2): connection errors, HTTP 408/409/429/5xx and provider retry headers. A transient failure after a response started (websocket error/close, terminated stream, 5xx/429) re-sends that one assistant request up to three times (1 s/2 s/4 s), never a tool; 0 disables both. Probes and summaries are single-shot.
providers.piNative.promptCacheDiagnostics	boolean	--	false	true	Emit metadata-only prompt-cache request fingerprints into run artifacts; never prompt text, tool arguments, cache keys, endpoints or credentials.
providers.piNative.transport	string	--	auto	sse	Preferred Pi provider transport: auto, sse, websocket, or websocket-cached. Providers without multiple transports ignore it.
runtime.compaction.contextWindowOverride	integer	--	auto-detected	128000	Persistent correction for inaccurate provider context-window metadata; learned overflow ceilings may lower it process-locally.
runtime.compaction.effort	string	--	auto	auto	Independent summary effort: auto disables thinking when supported, otherwise uses low or the lowest supported level; never inherits conversation effort.
runtime.compaction.enabled	boolean	--	true	true	Enables adaptive proactive compaction and one-shot reactive overflow recovery.
runtime.compaction.failureCooldownMs	integer	--	900000	900000	Suppress unchanged failed summaries for this many milliseconds (default 15 minutes); 0 disables suppression. Meaningful growth or summary-policy changes allow an earlier attempt.
runtime.compaction.fixedOverheadEnabled	boolean	--	true	true	Includes system instructions, tool schemas, and the current user turn in proactive request-size estimates.
runtime.compaction.keepRecentTokens	integer	--	adaptive by model	12800	Explicit recent-context retention override; omitted derives 10% of the effective context window, clamped to 4,000-20,000 tokens.
runtime.compaction.maxInputTokens	integer	--	200000	300000	Absolute proactive compaction ceiling in tokens; the trigger is the lowest of the ratio, the safety-headroom trigger and this value, so windows up to about 222,000 tokens are unaffected by the default. 0 disables it; other values must be 32,000-10,000,000. A soft trigger only: it never refuses a request or changes the usable window, and it is lifted when fixed request overhead plus summary, retained tail and minimum savings would not fit beneath it.
runtime.compaction.minSavingsTokens	integer	--	adaptive by model	12800	Minimum verified token reduction required for proactive compaction; omitted derives 10% of the effective window, clamped to 4,000-20,000. Reactive recovery accepts any positive reduction.
runtime.compaction.summaryMaxTokens	integer	--	adaptive by model	5120	Combined generated-summary text target; omitted derives 4% of the effective window, clamped to 2,000-12,000 tokens. Adaptive thinking may add a bounded 2,048-8,192-token allowance per request, subject to model/context limits.
runtime.compaction.triggerRatio	number	--	0.9	0.9	Fraction of the effective model context window used for the proactive trigger, additionally capped by adaptive safety headroom (10% of the window, 16,000-48,000 tokens).
runtime.effort	string	--	unset	medium	Route-specific effort forwarded through Pi to the selected provider. Doctor warns when a configured value falls outside the model’s advertised ladder but keeps turn-time handling permissive. Message text never changes effort.
runtime.fallbacks	array	--	[]	[{"model":"openai-codex:gpt-5.6-sol"},{"model":"anthropic:claude-sonnet-4-6","effort":"high"}]	Canonical ordered fallback routes. Omitted per-route effort means that provider’s default.
runtime.maxTurns	integer	--	unset	1	Configures maxTurns for the runtime section.
runtime.model	`string	object`	--	required	{"model":"openai-codex:gpt-6.1-sol","context1M":true}
runtime.retry.backoffMs	integer	--	1000	2000	Delay before the first same-model retry. Doubles on each further retry, capped by runtime.retry.maxBackoffMs.
runtime.retry.maxBackoffMs	integer	--	15000	30000	Ceiling for the doubling same-model retry delay.
runtime.retry.primaryAttempts	integer	--	2	3	Total attempts on runtime.model including the first, before the chain advances. Retries fire only for transient provider failures (overloaded, rate-limited, timeout, network, 5xx); context overflow and bad credentials advance immediately. Set 1 to disable.
runtime.session.continuity.enabled	boolean	--	true	true	Default-on durable conversation evidence and ConversationHistory retrieval in the current logical conversation. No memory backend is required. Retains sanitized user/assistant text, portable summaries and permitted tool text in the existing sidecar, at most 1 MiB per record within its shared 256 MiB payload budget. Pins never silently expire; capacity exhaustion reports degraded continuity. Disabled capture does not upgrade a v2 sidecar; existing v3 evidence stays on disk for doctor, stats and purge, but ConversationHistory is not registered while disabled. Disabling continuity leaves previously retained evidence and any pending purges at rest until continuity is re-enabled or storage is purged via session admin or restart —clear-sessions. That evidence is never served while a purge is pending. Enabling upgrades the tool-history sidecar to v3 in place; older binaries refuse it, so rollback requires a pre-upgrade backup. Set runtime.session.continuity.enabled=false to opt out with legacy capture/storage behavior. This does not add a ledger, paid upkeep or tool-result clearing; deterministic cold packs are controlled separately by packEnabled.
runtime.session.continuity.ledgerEnabled	boolean	--	false	true	Opt-in read-only reuse of a verified ledger, without buying upkeep. Upkeep also enables ledger reading. False by default; deterministic controls leave both switches off.
runtime.session.continuity.packEnabled	boolean	--	true	true	Default-on deterministic continuity evidence at cold canonical-history rebuilds. No model call. At most 8000 estimated tokens, further limited by target headroom; whole owner quotations, finalized tool evidence, latest summary and coverage/omission marker. Subordinate historical requests are not fresh approval. False keeps B1 recording and retrieval but injects no pack; continuity.enabled=false takes precedence. True native resumes get no additional pack. Per-message replay pays pack input tokens each turn. Legacy canonical messages lack exact record identities, so uncertain overlap is retained and reported. Native model-switch handoffs and subagents are unchanged.
runtime.session.continuity.upkeep.enabled	boolean	--	false	true	Opt-in boundary ledger upkeep. One no-tools completion on the turn’s effective primary conversation model or explicit extractionModel, no fallback; up to 128k bootstrap / 64k incremental estimated input, 4k output including reasoning, 60 seconds before first token. Only primary owned continuous cold seeds. Disabled by default in P3a.
runtime.session.continuity.upkeep.extractionModel	string	--	conversation model	openai:gpt-4o-mini	Explicit upkeep model override; otherwise the turn’s effective primary conversation model, captured before fallback. Switches also select extraction, including private-to-cloud evidence egress (up to about 128k retained tokens). Pin the previous runtime.model here to preserve fixed extraction. No fallback; budgets unchanged.
runtime.session.continuity.upkeep.requireRequestCap	boolean	--	false	true	Skip before dispatch when the adapter cannot emit a server output cap. Default false. Codex otherwise uses conservative visible-stream abort; hidden reasoning/quota cannot be client-capped. Reported overrun invalidates extraction.
runtime.session.idleTimeoutMs	integer	--	1800000	1800000	Configures session.idleTimeoutMs for the runtime section.
runtime.session.isolateProactive	boolean	--	false	true	Configures session.isolateProactive for the runtime section.
runtime.session.mode	string	--	continuous	continuous	Conversation session mode. Every durable-history user upgrades one way: the first durable turn rewrites canonical history as v3, with or without runtime.session.root or model switching. Stop all writers and back up canonical history and any native root first; rollback requires restoring the consistent backup.
runtime.session.modelSwitch.enabled	boolean	--	false	true	OFF-default Web-only durable model switching. Requires continuous sessions, runtime.session.root. Every model change, including a return to an earlier model, uses a structured handoff (outgoing summary, then the previously validated summary plus later history, then incoming summary); there is no summary-less model change and native session state is never reused across models. Only Web persisted user-message IDs authorize switches or new summary generations (at most two billed summary calls per switch per generation); No-ID turns on non-native conversations keep the original cold-replay path; no-ID turns on native conversations inherit the current model even when a different sticky model is declared; model-specific context-window overrides for that different model do not carry. Changing runtime.model does not move existing native conversations until a persisted Web message selects the new model. Pending messages are not queued or replayed. With this on, Web turns with a persisted message ID (including a new conversation’s first message) and no-ID wakes on native conversations use the primary model only: no configured fallback-chain backup or router re-attempt. Prepared requests resolve ordinary provider authentication on each actual call, including OAuth refresh; credentials are not lease-pinned. Other turns keep fallbacks.
runtime.session.rollover	string	--	none	none	Configures session.rollover for the runtime section.
runtime.session.rolloverNotice	boolean	--	false	true	Configures session.rolloverNotice for the runtime section.
runtime.session.rolloverTimezone	string	--	unset	example	Configures session.rolloverTimezone for the runtime section.
runtime.session.root	string	--	unset	example	Durable sessions root. The first durable turn after upgrading writes session journals, .pending-turns and the new canonical history format. This is a one-way upgrade: older binaries must not run against this root; rollback requires restoring a backup.
runtime.workspace	string	--	.	.	Configures workspace for the runtime section.
sandbox.denyWrite	string[]	--	unset	["example"]	Configures denyWrite for the sandbox section.
sandbox.fallback	string	--	fail-closed	fail-closed	Configures fallback for the sandbox section.
sandbox.mode	string	--	unset	native	Configures mode for the sandbox section.
sandbox.network.allowlist	string[]	--	unset	["example"]	Configures network.allowlist for the sandbox section.
sandbox.network.mode	string	--	none	none	Configures network.mode for the sandbox section.
sandbox.readableRoots	string[]	--	unset	["example"]	Configures readableRoots for the sandbox section.
sandbox.unsafeAllowHostProcess	boolean	--	false	true	Configures unsafeAllowHostProcess for the sandbox section.
sandbox.writableRoots	string[]	--	unset	["example"]	Configures writableRoots for the sandbox section.
slack.allowAllChannels	boolean	MONO_AGENT_SLACK_ALLOW_ALL_CHANNELS	false	true	Configures allowAllChannels for the slack section.
slack.allowedChannelIds	string[]	MONO_AGENT_SLACK_ALLOWED_CHANNEL_IDS	unset	["example"]	Configures allowedChannelIds for the slack section.
slack.appToken	string	MONO_AGENT_SLACK_APP_TOKEN	unset	env:MONO_AGENT_SLACK_APP_TOKEN	Secret value for slack.appToken; prefer the env override.
slack.botToken	string	MONO_AGENT_SLACK_BOT_TOKEN	unset	env:MONO_AGENT_SLACK_BOT_TOKEN	Secret value for slack.botToken; prefer the env override.
slack.botUserIds	string[]	MONO_AGENT_SLACK_BOT_USER_IDS	unset	["example"]	Configures botUserIds for the slack section.
slack.drainDeadlineMs	integer	MONO_AGENT_SLACK_DRAIN_DEADLINE_MS	unset	1	Configures drainDeadlineMs for the slack section.
slack.enabled	boolean	MONO_AGENT_SLACK_ENABLED	false	true	Enables the slack capability.
slack.heartbeatIntervalMs	integer	MONO_AGENT_SLACK_HEARTBEAT_INTERVAL_MS	unset	1	Configures heartbeatIntervalMs for the slack section.
slack.heartbeatTimeoutMs	integer	MONO_AGENT_SLACK_HEARTBEAT_TIMEOUT_MS	unset	1	Configures heartbeatTimeoutMs for the slack section.
slack.homeTab	object	--	unset	{"enabled":true,"headerText":"*Quick actions*","buttons":[{"actionId":"triage","label":"Triage","prompt":"Triage today's requests.","channelId":"C0123"}]}	JSON-only Slack App Home configuration; enabled is optional (default false), buttons is optional (default []), and there is no environment-variable form.
slack.mentionTextAliases	string[]	MONO_AGENT_SLACK_MENTION_TEXT_ALIASES	unset	["example"]	Configures mentionTextAliases for the slack section.
slack.reconnectInitialBackoffMs	integer	MONO_AGENT_SLACK_RECONNECT_INITIAL_BACKOFF_MS	unset	1	Configures reconnectInitialBackoffMs for the slack section.
slack.reconnectMaxBackoffMs	integer	MONO_AGENT_SLACK_RECONNECT_MAX_BACKOFF_MS	unset	1	Configures reconnectMaxBackoffMs for the slack section.
slack.reconnectStabilityMs	integer	MONO_AGENT_SLACK_RECONNECT_STABILITY_MS	unset	1	Configures reconnectStabilityMs for the slack section.
slack.reconnectStartupGraceMs	integer	MONO_AGENT_SLACK_RECONNECT_STARTUP_GRACE_MS	unset	1	Configures reconnectStartupGraceMs for the slack section.
slack.resolveChannelNames	boolean	MONO_AGENT_SLACK_RESOLVE_CHANNEL_NAMES	true	true	Resolves the channel’s name via conversations.info so the agent knows WHICH channel it is talking in, not just that it is in one. Requires channels:read (public) / groups:read (private); a missing scope degrades to the surface kind and id alone rather than failing turns.
slack.resolveUserNames	boolean	MONO_AGENT_SLACK_RESOLVE_USER_NAMES	true	true	Resolves the speaker’s display name and handle via users.info so the agent knows who is talking. Requires the users:read scope; a missing scope degrades to an unnamed speaker rather than failing turns.
slack.shortcuts	array	--	[]	[{"callbackId":"triage","prompt":"Prepare the daily support triage checklist.","channelId":"C0123"}]	JSON-only global/message Slack shortcut bindings; there is no environment-variable form.
slack.stripMentionText	boolean	MONO_AGENT_SLACK_STRIP_MENTION_TEXT	preserve	false	When unset, preserves one readable authenticated self-mention marker; true restores legacy full stripping and false keeps raw mention forms.
slack.threadContext.enabled	boolean	MONO_AGENT_SLACK_THREAD_CONTEXT_ENABLED	true	true	Sends what was said in the conversation before the agent was triggered as untrusted background context. Reads the thread for an in-thread trigger and recent channel history otherwise; needs a *:history scope.
slack.threadContext.includeBotMessages	boolean	MONO_AGENT_SLACK_THREAD_CONTEXT_INCLUDE_BOT_MESSAGES	true	true	Include other apps’ messages (a CI or alert bot) labelled as bots. The agent’s own posts are always excluded.
slack.threadContext.maxMessages	integer	MONO_AGENT_SLACK_THREAD_CONTEXT_MAX_MESSAGES	15	15	Messages of context sent per turn, newest kept. 0 disables the read. Capped at the harness’s own 30-message bound.
slack.threadContext.requestLimit	integer	MONO_AGENT_SLACK_THREAD_CONTEXT_REQUEST_LIMIT	15	15	Objects requested from Slack per read. Slack caps this at 15 for non-Marketplace apps; internal apps can raise it.
slack.threadContext.timeoutMs	integer	MONO_AGENT_SLACK_THREAD_CONTEXT_TIMEOUT_MS	4000	4000	Budget for the whole context phase, including name resolution. Exceeding it submits the turn with less context rather than delaying it. 0 bounds the phase only by the turn itself.
slack.unfurlLinks	boolean	MONO_AGENT_SLACK_UNFURL_LINKS	unset	false	Controls link previews on native Slack agent chat.postMessage requests. When unset, the request field is omitted and Slack’s current default behavior is preserved.
slack.unfurlMedia	boolean	MONO_AGENT_SLACK_UNFURL_MEDIA	unset	false	Controls media previews on native Slack agent chat.postMessage requests. When unset, the request field is omitted and Slack’s current default behavior is preserved.
subagents	object	--	{“enabled”:false}	{"enabled":true,"maxConcurrent":5,"instances":{"enabled":true,"maxPerConversation":8,"idleTtlMs":86400000,"maxTurns":60},"models":[{"name":"fable","model":"anthropic:claude-fable-5-1"},"openai-codex:gpt-6-astra"],"definitions":[{"name":"researcher","description":"Reads code and docs to answer a factual question about the codebase. Read-only.","prompt":"You are a codebase researcher. Answer with file:line citations. Never modify files.","allowedTools":["Read","Glob","Grep"]}]}	Subagent profiles the Agent tool can deploy, plus its caps. providerRetryBudgetMs (integer 0–180000 ms, default 120000) gives child assistant requests an outage recovery window; 0 restores legacy retries. piNative.piMaxRetries:0 disables both retry kinds. Interactive parent turns keep legacy policy. Bedrock retains AWS retry ownership. instances enables conversation-scoped persistent children when subagents are enabled and effective tool policy allows both Agent and AgentManage: root defaults to <artifacts.dir>/../subagents, maxPerConversation is 1–32 (default 8), idleTtlMs is 60000–604800000 (default 86400000), and maxTurns is 1–500 (default 60). Agent persist:true creates a child; AgentManage resumes or closes it. Persistent children automatically receive AskParent unless global/profile policy denies it; it durably saves a question and returns successful awaiting_reply. Answer through ordinary AgentManage in the same session. Disabling instances preserves stateless Agent. subagents.timeoutMs and definitions[].timeoutMs accept 1000–14400000 ms (up to four hours); a detached child’s actual timer is clamped to remaining process-job runtime less a settlement margin (10% for short jobs, at most 15 s), and the hard job deadline still wins. Full four-hour child turns require processJobs.maxRuntimeMs above four hours. commandTimeoutMs (positive integer milliseconds, default 1800000) is the foreground Bash/Exec ceiling for all children: detached children clamp it to their process job’s remaining runtime, and foreground children clamp it to their remaining turn time minus a settlement reserve (10% for short turns, at most 15 s). Child timers and parent aborts still stop commands sooner; interactive parent turns keep the 120 s default. models is an operator allow-list of model reference strings or {name?, model} objects; an omitted name uses the canonical reference. Explicit names must be lowercase kebab-case (1-40 characters), unique, and distinct from profile names and general-purpose; duplicate models are rejected. Agent accepts model from that list and effort on every call shape. Each value resolves as call-time override, profile pin, parent effective value, then base/runtime default. Disabled unless enabled is true, in which case Agent must also appear in tools.allowedTools. Each definition needs exactly one of prompt or promptPath; omitted allowedTools means a read-only default set, and the ”*” wildcard is rejected. The agent may also author a specialized subagent at call time unless inline.enabled is false; inline.allowedTools caps what an authored subagent may request, defaulting to the parent agent’s own built-ins. maxConcurrent is a ceiling, not a scheduling guarantee: Agent and read-only calls can overlap even when stateful tools are offered, but an invoked stateful, MCP, or unknown call is an exclusive FIFO barrier. Background jobs and detached children release admission on their started receipt.
telegram.allowAllChats	boolean	MONO_AGENT_TELEGRAM_ALLOW_ALL_CHATS	false	true	Configures allowAllChats for the telegram section.
telegram.allowedChatIds	string[]	MONO_AGENT_TELEGRAM_ALLOWED_CHAT_IDS	unset	["example"]	Configures allowedChatIds for the telegram section.
telegram.apiRoot	string	MONO_AGENT_TELEGRAM_API_ROOT	unset	example	Configures apiRoot for the telegram section.
telegram.attachments.downloadTimeoutMs	integer	MONO_AGENT_TELEGRAM_ATTACHMENT_DOWNLOAD_TIMEOUT_MS	unset	1	Configures attachments.downloadTimeoutMs for the telegram section.
telegram.attachments.maxBytes	integer	MONO_AGENT_TELEGRAM_ATTACHMENT_MAX_BYTES	unset	1	Configures attachments.maxBytes for the telegram section.
telegram.attachments.maxUploadBytes	integer	MONO_AGENT_TELEGRAM_UPLOAD_MAX_BYTES	unset	1	Configures attachments.maxUploadBytes for the telegram section.
telegram.botToken	string	MONO_AGENT_TELEGRAM_BOT_TOKEN	unset	env:MONO_AGENT_TELEGRAM_BOT_TOKEN	Secret value for telegram.botToken; prefer the env override.
telegram.commands	array	--	[]	[{"command":"status","prompt":"Report current status."}]	Telegram command definitions handled by the Telegram adapter.
telegram.enabled	boolean	MONO_AGENT_TELEGRAM_ENABLED	false	true	Enables the telegram capability.
telegram.groupMode	string	MONO_AGENT_TELEGRAM_GROUP_MODE	any	mention	Group-message trigger rule: any runs every allowed group message; mention runs only native @mentions of the bot and replies to its messages; listen triggers like mention but gives the next turn the unaddressed messages since the bot last answered as untrusted background context. Direct chats and commands are unaffected.
telegram.pollWatchdogMs	integer	MONO_AGENT_TELEGRAM_POLL_WATCHDOG_MS	unset	1	Configures pollWatchdogMs for the telegram section.
telegram.projects.enabled	boolean	--	false	true	Mirror each forum topic the bot sees (and a forum’s General conversation) one way into a web-console project, record observations under .mono-agent/telegram-topics-v1/, inject the project’s shared context into every turn there, and add projectId to the Telegram send tools. Needs a running web console (mono-agent web run); project tools on Telegram turns stay gated by tools.allowedTools.
telegram.quietHours	object	--	unset	{"timezone":"UTC","start":"22:00","end":"07:00"}	Quiet-hours rules for Telegram notifications.
telegram.reactions	object	MONO_AGENT_TELEGRAM_REACTIONS	unset	{"working":true,"done":true,"error":true}	Telegram lifecycle reactions. The env override is boolean and toggles all states.
telegram.sendTools.pathScope	string	--	unset	run-output	Confine Telegram path uploads to the current run output directory.
telegram.sendTools.scope	string	--	unset	producing-conversation	Bind Telegram send tools to the chat, or forum topic, that produced the current run.
telegram.stripMentionText	boolean	MONO_AGENT_TELEGRAM_STRIP_MENTION_TEXT	true	true	Removes matching native @mentions from responder text in mention mode; replies without a mention are unchanged.
telegram.topics	array	--	[]	[{"chatId":"-1001234567890","topicId":12,"groupMode":"any"}]	Per-forum-topic trigger overrides for allowlisted chats: groupMode is inherit (default), any, mention, or listen. The General topic follows telegram.groupMode; entries never widen the chat allowlist.
telegram.transcription.endpoint	string	MONO_AGENT_TELEGRAM_TRANSCRIPTION_ENDPOINT	unset	example	Configures transcription.endpoint for the telegram section.
telegram.transcription.language	string	MONO_AGENT_TELEGRAM_TRANSCRIPTION_LANGUAGE	unset	example	Configures transcription.language for the telegram section.
telegram.transcription.model	string	MONO_AGENT_TELEGRAM_TRANSCRIPTION_MODEL	unset	example	Configures transcription.model for the telegram section.
telegram.transcription.timeoutMs	integer	MONO_AGENT_TELEGRAM_TRANSCRIPTION_TIMEOUT_MS	unset	1	Configures transcription.timeoutMs for the telegram section.
telegram.transport.ipFamily	integer	MONO_AGENT_TELEGRAM_IP_FAMILY	unset	example	Configures transport.ipFamily for the telegram section.
tools.allowedTools	string[]	--	[”*“]	["Read","Grep"]	Configures allowedTools for the tools section.
tools.computerUse.backend	string	--	unset	cua-driver	Opt into local desktop control using “cua-driver” (only backend). Absent means disabled; MCP tools bypass built-in allow/deny policy. See Computer use for safety and permission boundaries.
tools.computerUse.command	string	--	unset	/opt/tools/cua-driver	Optional executable name or path. Relative paths resolve against the config workspace. Otherwise resolve PATH, then platform installation locations; doctor reports the selected path.
tools.continuationServers	string[]	--	unset	["example"]	Configures continuationServers for the tools section.
tools.conversationSearch.datedSnippets	boolean	--	false	true	Opt in to dated SearchConversations message evidence on verified owner web turns. Requires local BuJo memory; disabled by default. Telegram and the console UI keep legacy search. No new index or stored history fields.
tools.disallowedTools	string[]	--	[]	["Read","Grep"]	Configures disallowedTools for the tools section.
tools.filesystem.readableRoots	string[]	--	unset	["/srv/shared/reference"]	Extra roots for managed Read, Glob, and Grep while sandbox.mode is off. Access requires lexical and realpath containment; native sandbox roots remain authoritative when sandboxing is enabled.
tools.filesystem.writableRoots	string[]	--	unset	["/srv/shared/output"]	Extra roots for managed Write and Edit while sandbox.mode is off; writable roots are also readable. Access requires lexical and realpath containment; native sandbox roots remain authoritative when sandboxing is enabled.
tools.mcpCallMaxTotalTimeoutMs	integer	--	2700000	2700000	Configures mcpCallMaxTotalTimeoutMs for the tools section.
tools.mcpCallTimeoutMs	integer	--	120000	120000	Configures mcpCallTimeoutMs for the tools section.
tools.mcpConfigPath	string	--	unset	example	Configures mcpConfigPath for the tools section.
tools.mcpRequestContextServers	string[]	--	[]	["transcribe"]	Configured stdio MCP server names that receive trusted per-request conversation, run, output-directory, and scoped progress context.
tools.web.coordination	string	--	process	process	Host mode shares web request budgets and cooldowns across participating local mono-agent processes; process preserves isolated coordination.
tools.web.fetch.browserCommand	string	--	agent-browser	agent-browser	Direct executable name or path for agent-browser; shell fragments are not evaluated.
tools.web.fetch.parallel.apiKeyEnv	string	--	unset	example	Secret value for tools.web.fetch.parallel.apiKeyEnv; prefer the env override.
tools.web.fetch.provider	`string	string[]`	--	local	local
tools.web.fetch.render	string	--	never	never	Browser-render capability for sparse JavaScript pages. never forces every call to static extraction; auto permits an isolated agent-browser session when needed.
tools.web.search.backend	`string	string[]`	--	[“parallel”,“ollama”]	parallel,ollama
tools.web.search.codex.model	string	--	gpt-5.6-luna	gpt-5.6-luna	Codex app-server model used for ChatGPT-subscription web search. The signed-in account must expose both this model and web search; default gpt-5.6-luna.
tools.web.search.endpoint	string	--	unset	example	Compatibility alias for tools.web.search.searxng.endpoint. Existing configurations remain valid; new configurations should use the provider-specific block.
tools.web.search.maxRequestsPerRun	integer	--	4	4	Hard ceiling from 1 to 20 on answered provider searches in one logical runtime run, including empty answers; failures are refunded. Cache hits, coalesced followers, cooldown skips, and quota skips consume no request. A separate dispatch ceiling is four times this limit. Default 4.
tools.web.search.ollama.apiKeyEnv	string	--	unset	OLLAMA_API_KEY	Environment variable containing the hosted Ollama API key. Required only for exact https://ollama.com and rejected for every other origin; the key value is never stored in JSON.
tools.web.search.ollama.baseUrl	string	--	http://127.0.0.1:11434	https://ollama.com	Ollama origin. Defaults to local http://127.0.0.1:11434 when Ollama is selected, including in a chain. Hosted search is exactly https://ollama.com; custom public origins require HTTPS and trustPublicUrl=true.
tools.web.search.ollama.trustPublicUrl	boolean	--	false	true	Explicit acknowledgement for an HTTPS custom public Ollama origin. It never permits sending hosted credentials to that origin.
tools.web.search.parallel.apiKeyEnv	string	--	unset	example	Secret value for tools.web.search.parallel.apiKeyEnv; prefer the env override.
tools.web.search.searxng.endpoint	string	--	unset	http://127.0.0.1:8088	Optional unauthenticated loopback HTTP SearXNG base URL. Remote HTTPS, credentials, query strings, and fragments are rejected.
traceability.globalDiscovery	boolean	--	true	true	Configures globalDiscovery for the traceability section.
traceability.heartbeatMs	integer	--	10000	10000	Configures heartbeatMs for the traceability section.
traceability.registryDir	string	--	.mono-agent/trace-sources	.mono-agent/trace-sources	Configures registryDir for the traceability section.
traceability.sourceId	string	--	unset	my-agent	Configures sourceId for the traceability section.
traceability.sourceLabel	string	--	unset	My Agent	Configures sourceLabel for the traceability section.
traceability.staleAfterMs	integer	--	30000	30000	Configures staleAfterMs for the traceability section.
tui.allowNonLoopback	boolean	MONO_AGENT_TUI_ALLOW_NON_LOOPBACK	false	true	Configures allowNonLoopback for the tui section.
tui.apiKey	string	MONO_AGENT_TUI_API_KEY	unset	example	Secret value for tui.apiKey; prefer the env override.
tui.basePath	string	MONO_AGENT_TUI_BASE_PATH	/gui	/gui	Configures basePath for the tui section.
tui.enabled	boolean	MONO_AGENT_TUI_ENABLED	true	true	Enables the tui capability.
tui.host	string	MONO_AGENT_TUI_HOST	127.0.0.1	127.0.0.1	Configures host for the tui section.
tui.port	integer	MONO_AGENT_TUI_PORT	0	0	Configures port for the tui section.
tui.requestToolEnvironment.allowedKeys	string[]	MONO_AGENT_TUI_REQUEST_TOOL_ENVIRONMENT_ALLOWED_KEYS	[]	["MULTICA_TOKEN","MULTICA_TASK_ID"]	Explicit environment-variable names an ACP request may pass to Bash, Exec, and nested subagents for one turn. Disabled when empty; dangerous process-loader, shell-startup, home, temp, and PATH keys are rejected.
tui.requestToolEnvironment.allowPathPrepend	boolean	MONO_AGENT_TUI_REQUEST_TOOL_ENVIRONMENT_ALLOW_PATH_PREPEND	false	false	Allows an ACP request to prepend up to four absolute directories to process-tool PATH for one turn. The caller cannot replace PATH.
webhook.allowNonLoopback	boolean	MONO_AGENT_WEBHOOK_ALLOW_NON_LOOPBACK	unset	true	Configures allowNonLoopback for the webhook section.
webhook.apiKey	string	MONO_AGENT_WEBHOOK_API_KEY	unset	set-via-MONO_AGENT_WEBHOOK_API_KEY	Secret value for webhook.apiKey; prefer the env override.
webhook.defaultMode	string	MONO_AGENT_WEBHOOK_DEFAULT_MODE	sync	sync	Configures defaultMode for the webhook section.
webhook.dir	string	MONO_AGENT_WEBHOOK_DIR	webhook	webhook	Configures dir for the webhook section.
webhook.effort	string	MONO_AGENT_WEBHOOK_EFFORT	unset	example	Configures effort for the webhook section.
webhook.enabled	boolean	MONO_AGENT_WEBHOOK_ENABLED	false	true	Enables the webhook capability.
webhook.endpoints	array	MONO_AGENT_WEBHOOK_ENDPOINTS_JSON	[]	[{"name":"triage","path":"/webhook/triage","prompt":"Triage this payload."}]	Named webhook endpoints with per-endpoint prompt, model/effort, and maxRunMs overrides.
webhook.host	string	MONO_AGENT_WEBHOOK_HOST	127.0.0.1	127.0.0.1	Configures host for the webhook section.
webhook.maxAttachmentBytes	integer	MONO_AGENT_WEBHOOK_MAX_ATTACHMENT_BYTES	20971520	20971520	Configures maxAttachmentBytes for the webhook section.
webhook.maxRunMs	integer	MONO_AGENT_WEBHOOK_MAX_RUN_MS	unset	1	Configures maxRunMs for the webhook section.
webhook.maxStoredRequests	integer	MONO_AGENT_WEBHOOK_MAX_STORED_REQUESTS	100	100	Configures maxStoredRequests for the webhook section.
webhook.model	string	MONO_AGENT_WEBHOOK_MODEL	unset	example	Configures model for the webhook section.
webhook.notify	boolean	MONO_AGENT_WEBHOOK_NOTIFY	unset	example	Configures notify for the webhook section.
webhook.notifyConversationId	string	MONO_AGENT_WEBHOOK_NOTIFY_CONVERSATION_ID	unset	example	Configures notifyConversationId for the webhook section.
webhook.path	string	MONO_AGENT_WEBHOOK_PATH	/webhook/invoke	/webhook/invoke	Configures path for the webhook section.
webhook.port	integer	MONO_AGENT_WEBHOOK_PORT	0	0	Configures port for the webhook section.
webhook.prompt	string	MONO_AGENT_WEBHOOK_PROMPT	unset	example	Configures prompt for the webhook section.
webhook.retentionMs	integer	MONO_AGENT_WEBHOOK_RETENTION_MS	300000	300000	Configures retentionMs for the webhook section.

Plugin channels

channels.plugins[] entries are intentionally open at config: the plugin package owns that nested payload. The host validates the plugin envelope (package, optional id, optional label, and config) and each loaded plugin reports its own config warnings.