On this page
On this page
CLI commands
Config
Non-interactive helpers for openclaw.json: get/set/patch/unset a value by path, print the schema, validate, or print the active file path. Run openclaw config with no subcommand to open the same guided wizard as openclaw configure.
config set, config patch, and config unset require the local Gateway to be
stopped. These commands hold exclusive state ownership while publishing the file
and its database metadata. Stop the Gateway through its service owner, wait for
it to exit, then rerun the command.
Externally managed config
Set OPENCLAW_CONFIG_READONLY=1 in the environment of both the Gateway and any
OpenClaw CLI processes when a deployment system manages your config:
For a service or container, set the variable in its service environment or
container definition. This is a host-environment switch, not an openclaw.json
field. Do not set OPENCLAW_CONFIG_READONLY in config env or env.vars.
Entries in env.vars are ignored, including differently cased spellings; flat
env keys are not valid configuration. Config reload cannot enable, disable,
or change the host-selected read-only mode. Only the host value 1 enables
this switch. Existing OPENCLAW_NIX_MODE behavior is unchanged.
Config writes are blocked, including setup, onboarding, Doctor config repairs, plugin
install/update/uninstall/enable/disable, and mutating openclaw update flows.
Startup-derived defaults stay runtime-only. Change the config through your
external deployment system, then let the Gateway reload it or restart the Gateway
as needed. Runtime state still needs a writable OPENCLAW_STATE_DIR.
Doctor's --fix --non-interactive pass, including the official Docker image's
startup pass, still repairs writable SQLite, session, and plugin state in this
mode. Pending config repairs are printed as a redacted merge patch to apply in
your external deployment source; the mounted config and its includes stay
unchanged. Valid config permits Gateway startup even with optional repairs
pending. If legacy or invalid config prevents startup, Doctor exits nonzero and
names the required edits. State schema migration runs before config repair so
plugin and session migrations can use the current schema; keep a matched
pre-upgrade state backup when reverting to an older image.
OPENCLAW_CONFIG_READONLY=1 uses generic externally managed config messages and
does not enable Nix-specific installation or service behavior. OPENCLAW_NIX_MODE=1
continues to imply immutable config, even if OPENCLAW_CONFIG_READONLY is unset or
0. For Nix installs, edit the Nix source instead; see Nix.
Root options
--section <section>stringRepeatable guided-setup section filter when you run openclaw config without a subcommand.
Guided sections: workspace, model, web, gateway, daemon, channels, plugins, skills, health.
Examples
Paths
Dot or bracket notation. Quote bracket paths in shell examples so zsh does not glob-expand [0]:
Prefer agents.entries.<id> paths for agent edits. The legacy agents.list[0]
syntax and whole-list inputs still work with set, patch, and unset; writes
persist the current keyed roster. Indexed edits use the current roster order.
Within a batch, a submitted list keeps its order across subsequent keyed edits,
including when agent IDs are numeric strings. Existing roster-deletion and
$include ownership protections still apply.
When a legacy roster expands a single-agent installation in the root config file,
writes retire its default marker and preserve the existing agent's responsibilities
with explicit owners. An explicitly authored ownership: "explicit" cannot be
combined with a legacy default: true marker.
For root-file writes, changing session.store clears a copied
agents.defaults.sessionStore.agentId because that owner belongs to the previous
store. To assign the destination store's owner, set that owner path explicitly in
the same batch. Unrelated writes preserve the owner, including when session.store
is unset and per-agent default stores apply. A committed write that clears the
owner prints a warning naming the key and the store change.
If an older version already removed the owner, Doctor checks the retained config
backups and offers to restore the most recent owner with the same authored
session.store value. Restoration
requires interactive confirmation because the removal might have been intentional;
unattended Doctor runs show the recovery command instead. If no usable backup
remains, the legacy-session finding names agents.defaults.sessionStore.agentId
so you can assign the owner explicitly.
config get
Reads a value from the redacted config snapshot (secrets never print). --json prints the same redacted value as JSON; otherwise strings/numbers/booleans print bare and objects/arrays print as formatted JSON.
Pass exactly one config path. Extra arguments, including an empty quoted argument (""),
are rejected; they do not suppress validation of later options.
A schema-valid but unset authorable path explains that the runtime default applies and suggests
openclaw config set. For automatically managed metadata, such as meta.lastTouchedVersion,
meta.migrations.modelPolicyAllowlist, and meta.migrations.utilityModelSeparation, the message
instead explains that OpenClaw manages the values on config writes. Reading their unset parent
paths (meta or meta.migrations) gives the same explanation, not a refused set command.
An unknown path suggests openclaw config schema. With --json, unset and unknown paths use
the standard CLI JSON failure envelope on stdout and exit with status 1.
Without --json, diagnostics remain on stderr.
Nested paths inside open-ended parameter bags, such as agents.defaults.params.custom.nested,
are schema-valid even before they are set. This does not confirm that a provider supports the parameter.
Explicit null, false, 0, and empty strings remain readable values in both modes;
--json preserves their types. Optional fields with no runtime value are reported as unset.
config file
Prints the active config file path, resolved from OPENCLAW_CONFIG_PATH or the default location. The path names a regular file, not a symlink; see Write safety.
With --json, stdout contains an object with the resolved path under path.
config schema
Prints the generated JSON schema for openclaw.json to stdout.
What it includes
- The current root config schema, plus a root
$schemastring field for editor tooling. - Field
title/descriptiondocs metadata used by the Control UI. - Nested object, wildcard (
*), and array-item ([]) nodes inherit the sametitle/descriptionmetadata when matching field docs exist. anyOf/oneOf/allOfbranches inherit the same docs metadata too.- Best-effort live plugin + channel schema metadata when runtime manifests can be loaded.
- A clean fallback schema even when the current config is invalid.
The schema is JSON in both modes. --json is accepted as the explicit
machine-output spelling and keeps stdout reserved for the schema document.
config validate
Schema refusals from config set, config patch, and config unset explain the affected setting and confirm that no settings were saved. Correct the reported value or use openclaw config schema to inspect supported settings, then retry. These refusals still exit with status 1. Explicit validation reports settings that need correction without changing the file; config validate --json retains its valid: false, error, and issues fields for scripts.
Config read failures name the file being read and preserve the underlying error. Resolve the reported file-access or runtime problem, then retry; a read failure alone does not mean the settings need repair.
Human validation diagnostics quote literal record keys, such as agents.defaults.models["provider/model.v1"].alias, instead of displaying the dot inside a key as nested traversal. Numeric array positions use brackets, such as agents.entries.main.skills[0]. The issues[].path field in config validate --json keeps its existing dot-joined representation.
Validates the current config against the active schema without starting the gateway. It also checks provider/source compatibility for every registry-declared SecretRef, including disabled plugin or channel configuration. This strict command can report an inactive mismatch that does not block normal Gateway startup, where SecretRef resolution remains limited to effectively active surfaces.
After schema validation, it checks every configured manual exec provider's command path using the same non-executing trust checks as startup: file presence, symlinks, trusted directories, permissions, ownership, and Windows ACL availability. config set, config patch, and config unset apply these checks only to providers changed or referenced by the operation, including during dry runs. Replacing the secrets or secrets.providers collection checks every remaining provider. An unrelated inactive provider does not block targeted repairs or removal of that provider.
Path validation does not execute providers or verify their output. Passing it does not guarantee successful secret resolution; exec dry runs require --allow-exec to test that separately.
Provider and runtime params bags are intentionally typed as
Record<string, unknown> because their owners define the supported keys and
values. openclaw config validate can validate the container and overall
config shape, but it cannot type-check provider-specific parameter names or
values. Passing validation does not prove that a param is supported; consult
the provider docs and verify behavior on the selected runtime and provider.
Values
Values parse as JSON5 when possible; otherwise they are treated as raw strings. Use --strict-json to require standard JSON with no string fallback (JSON5-only syntax such as comments, trailing commas, or unquoted keys is then rejected). --json is a legacy alias for --strict-json on config set.
For structured values that are awkward to quote in your shell, put a config-shaped JSON5 object in a file and use config patch --file <path> --dry-run. The file contains config keys and their values, not a bare array.
config get <path> --json prints the redacted value as JSON instead of terminal-formatted text.
When a model uses string shorthand, setting its fallbacks or a supported tool-model timeoutMs preserves that string as primary. This also applies to chat /config set. Setting primary explicitly replaces the primary, and setting the whole model still replaces the whole value.
When a write changes agents.defaults.model or a per-agent agents.entries.*.model, OpenClaw resolves each changed primary or fallback through the configured catalogs and the selected provider's model resolver before writing. Provider-supported exact provider/model pins are accepted even when absent from the curated picker; validation does not replace the selected model. Unknown model references are rejected without changing the active config. Refusals for newly authored per-agent references include the entered value and resolver reason; inherited values and environment-expanded details remain redacted. Run openclaw models list to browse the picker, or check the provider's documentation for an exact model ID. Successful validation does not prove that your account can call the model. openclaw models set is deliberately more permissive for the same setting: it saves a model the local catalog cannot confirm and prints a warning instead of rejecting the write.
Use --merge when adding entries to those maps:
Use --replace only when the provided value should intentionally become the complete target value.
Conditional writes
Use a conditional expectation when automation must update one authored path only if it has not changed since the caller last observed it:
--expect-current-json <json> uses strict JSON and compares the value by JSON type and structure.
null is an authored value, so it does not satisfy --expect-current-absent. The comparison uses
the effective authored config after includes and environment substitution, before runtime defaults
are applied.
If the expectation does not match, no settings are saved. Read the current config and review the expected value before retrying; repeating the same mismatched expectation will not succeed.
The two expectation flags are mutually exclusive. They apply only to a single config set
operation, require a direct non-redirected config path, and cannot be combined with batch mode or
--dry-run. If input or roster resolution would write a different path than the caller requested,
such as a sibling *Ref path, the command exits with status 1 instead of retargeting the
expectation. A mismatch exits with status 1, writes nothing, and does not print either the expected
or current value. OpenClaw's config snapshot guard still rejects a later race between the
expectation check and the final file replacement.
config set modes
Value mode
SecretRef builder mode
Provider builder mode
Targets secrets.providers.<alias> paths only:
Batch mode
Batch files are limited to 8 MiB.
Batch parsing always uses the batch payload (--batch-json/--batch-file) as the source of truth; --strict-json / --json do not change batch parsing behavior.
Supplying either batch option selects batch mode. Empty or whitespace-only values are rejected; omit both options to use positional <path> <value> mode.
--batch-file and config patch --file use the exact file path you provide, including leading or trailing spaces. Quote paths that contain spaces in your shell.
Batch assignments apply in order, then validation checks the final config. A SecretRef replaced by a later assignment is not resolved or counted in dry-run output, even with --allow-exec. Providers that remain in a changed provider collection still receive command-path trust checks.
JSON path/value mode also works for SecretRefs and providers directly:
Provider builder flags
Provider builder targets must use secrets.providers.<alias> as the path.
Common flags
--provider-source <env|file|exec|store>--provider-timeout-ms <ms>(file,exec)
Env provider (--provider-source env)
--provider-allowlist <ENV_VAR>(repeatable)
File provider (--provider-source file)
--provider-path <path>(required)--provider-mode <singleValue|json>--provider-max-bytes <bytes>
Exec provider (--provider-source exec)
--provider-command <path>(required)--provider-arg <arg>(repeatable); each occurrence preserves one literal argument, including an empty string or surrounding whitespace. Quote these values in your shell.--provider-no-output-timeout-ms <ms>--provider-max-output-bytes <bytes>--provider-json-only--provider-env <KEY=VALUE>(repeatable)--provider-pass-env <ENV_VAR>(repeatable)--provider-trusted-dir <path>(repeatable)
Hardened exec provider example:
config patch
Paste or pipe a config-shaped JSON5 patch instead of running many path-based config set commands. Objects merge recursively; arrays and scalar values replace the target; null deletes the target path.
Patch files are limited to 8 MiB. Piped --stdin patches are limited to 1 MiB.
Pipe a patch over stdin for remote setup scripts:
Example patch:
The runtime pin makes this an embedded OpenClaw recipe. A valid fastMode
value is a portable typed runtime control and does not choose OpenClaw by
itself.
Use --replace-path <path> when one object or array must become exactly the provided value instead of being recursively patched:
--dry-run runs schema and SecretRef resolvability checks without writing. Exec-backed SecretRefs are skipped by default during dry-run; add --allow-exec when you intentionally want dry-run to execute provider commands.
Dry run
--dry-run simulates a change without writing openclaw.json. Available on config set, config patch, and config unset. Which checks run depends on the input mode. Value mode (config set <path> <value> without --strict-json) skips the full schema pass and the ordinary SecretRef resolvability scan. Policy, provider, and model-reference checks can still run. When no checks apply, value mode reports Dry run successful even for a value the real write rejects. Use --strict-json (or config patch --file --dry-run) when you need schema validation.
For config patch and config unset, --json requires --dry-run. Using --json without --dry-run returns the standard CLI JSON failure envelope on stdout, keeps diagnostics on stderr, and exits with status 1.
Dry-run behavior
- Value mode (a plain
<value>without--strict-json): skips the full schema pass and ordinary SecretRef resolvability scan. Policy, provider, and model-reference checks can still run. When no checks apply, the CLI printsDry run note: value mode does not run schema/resolvability checksand can succeed even when the real write would fail schema validation. - Builder mode: runs SecretRef resolvability checks for changed refs/providers. A SecretRef builder target outside the registered config secret paths also runs full schema validation, so an unsupported path fails instead of reporting a successful preview.
- JSON mode (
--strict-json,--json, or batch mode): runs schema validation plus SecretRef resolvability checks. - Policy validation runs against the full post-change config, so parent-object writes (for example setting
hooksas an object) cannot bypass unsupported-surface validation. - Exec command-path trust checks run without executing providers. Exec SecretRef resolvability checks are skipped by default to avoid command side effects; pass
--allow-execto opt in (this may execute provider commands).--allow-execis dry-run only and errors without--dry-run.
--dry-run --json fields
ok: whether dry-run passedoperations: number of assignments evaluatedchecks: whether schema/resolvability checks ranchecks.resolvabilityComplete: whether resolvability checks ran to completion (false when exec refs are skipped)refsChecked: number of refs actually resolved during dry-runskippedExecRefs: number of exec refs skipped because--allow-execwas not seterrors: structured failures whenok=false; each carries akindofmissing-path,schema,resolvability,model, orconflict(conflictmeans the write was declined because its config snapshot, target, or conditional expectation no longer matched; follow the message before retrying)
JSON output shape
Success example
Failure example
If dry-run fails
config schema validation failed: your post-change config shape is invalid; fix the path/value or provider/ref object shape.Config policy validation failed: unsupported SecretRef usage: move that credential back to plaintext/string input; keep SecretRefs on supported surfaces only.SecretRef assignment(s) could not be resolved: the referenced provider/ref cannot currently resolve (missing env/store name, invalid file pointer, exec provider failure, or provider/source mismatch).model reference validation failed: a changed text-model primary or fallback is unknown; runopenclaw models listand choose an available model.Dry run note: skipped <n> exec SecretRef resolvability check(s): rerun with--allow-execif you need exec resolvability validation.- For batch mode, fix failing entries and rerun
--dry-runbefore writing.
Applying changes
After every successful config set / config patch / config unset, the CLI prints one of three hints so you know whether the gateway needs a restart:
| Hint | Meaning |
|---|---|
Restart the gateway to apply. |
The changed path needs a full restart, or passive reload is disabled. |
Change will apply without restarting the gateway. |
Hot reload picks it up automatically. |
No gateway restart needed. |
Nothing runtime-relevant changed. |
Plugin entry changes use the same reload planner as other settings. In the default
hybrid mode, ordinary plugins.entries.<id> edits replace the affected plugin
instance automatically. A plugin's narrower restart policy or
gateway.reload.mode: "off" can still require a Gateway restart. The CLI hint
describes expected application; it is not a receipt from a running Gateway.
See Config hot reload.
Successful config set or config unset operations that produce no effective config diff print No change and leave the JSON5 file byte-for-byte untouched. A config unset target that is absent from the authored config exits with status 1 and also leaves the file untouched. Setting an absent key to a value equal to its runtime default is still an authored change and persists the explicit value.
Write safety
openclaw config set and other OpenClaw-owned config writers validate the full post-change config before committing it to disk. If the new payload fails schema validation or looks like a destructive clobber, the active config is left alone and the rejected payload is saved beside it as openclaw.json.rejected.*.
If staging a config save fails, the existing root or include backup ring is left untouched. OpenClaw prepares backup contents without blocking unrelated Gateway requests. If a copy fallback removes the file before a conflict, OpenClaw restores the original when it still owns the missing destination. Otherwise, the error reports partial publication and the backup location to inspect before another save.
If the file is saved but later processing fails, the error names the written file and reports whether the write was rolled back. This can name an included file when that file owns the edited setting. If rollback did not happen or could not be confirmed, inspect the named file and the active config before retrying.
OpenClaw-owned writes that change config reserialize JSON5 as standard JSON. When the source contains comments, the writer warns immediately before removing them; use a direct editor when preserving comments matters.
Prefer CLI writes for small edits:
If a write is rejected, inspect the saved payload and fix the full config shape:
Direct editor writes are still allowed, but the running Gateway treats them as untrusted until they validate. Startup validates config without rewriting legacy keys. Invalid direct edits stop startup; hot reload skips invalid edits without rewriting openclaw.json. Run openclaw doctor --fix for legacy-key repair, prefixed/clobbered config, or last-known-good recovery. See Gateway troubleshooting.
Ordinary recovery can restore an eligible, valid current backup verbatim. Backups that need legacy transformations must be recovered through Doctor. Plugin schema changes or minHostVersion skew stay loud instead of rolling back unrelated user settings such as models, providers, auth profiles, channels, gateway exposure, tools, memory, browser, or cron config.
Repair loop
After openclaw config validate passes, use the local TUI to have an embedded agent compare the active config against the docs while you validate each change from the same terminal:
Inside the TUI, a leading ! runs a literal local shell command (after a one-time per-session confirmation prompt):
Compare with docs
Ask the agent to compare your current config with the relevant docs page and suggest the smallest fix.
Apply targeted edits
Apply targeted edits with openclaw config set or openclaw configure.
Re-validate
Rerun openclaw config validate after each change.
Doctor for runtime issues
If validation passes but the runtime is still unhealthy, run openclaw doctor or openclaw doctor --fix for migration and repair help.
Related
- CLI reference
- Configuration
openclaw configure— guided editor for the same settings
