NylorunDocsBeta
RunConfigure

CLI

Every nylorun and create-agent command.

Two tools. nylorun runs local Tenants and is the client of a Tenant's APIs, local or remote. npm create @nylorun/agent writes a project.

npx nylorun

npx nylorun <command>

The installation's lifecycle, and writing, checking, and saving agent manifests, are top-level commands. Every other command is in a group named after what it manages, and each group has its own --help. up and down are aliases of start and stop.

CommandFlagsNotes
start / up--tenant <name>, --no-link, --no-studio, --allow-downgrade, --restate-ui, --studio-embed-origin <origin> (repeats), --studio-embed-origin-resetCreates the Tenant and the project link on first run and prints the Runtime and Studio URLs; it opens no browser. --no-link starts a Tenant without linking the directory. Exit 5 if this nylorun pins an older Runtime than the Tenant last ran. Passes NYLORUN_TUNING (Host tuning), the OTEL_* variables, and NYLORUN_PUBLIC_URL (default http://localhost:<port>, the base of webhook URLs) from its shell to the containers. It does not pass NYLORUN_WORKLOAD_TOKEN_FILE or NYLORUN_WORKLOAD_TOKEN_ORIGINS, so a local Tenant cannot use the workload token.
stop / down--tenant <name> or --allKeeps volumes. --all cannot combine with --tenant (exit 2).
status--tenant <name>, --jsonThe Tenant, its id, services, endpoints, and Runtime health.
studio--tenant <name>, --no-openSigns a browser in to Studio, starting the Tenant if it is stopped. --no-open prints a single-use login URL, valid for two minutes.
doctor--jsonNode, Docker, Compose v2, and Tenant health.
doctor sandbox--tenant <name>, --jsonThe sandbox backend the Tenant's Host offers and, with pods enabled, a pods line: agent-sandbox or Agent Substrate, the context, the namespace, and whether it is ready.
--versionPrints the version.
init <path>--description <text>, --exampleWrites <path>/AGENT.md, the smallest agent manifest, named after the path's last segment. --example adds an HTTP tool, mcp.json, and a trigger, with a README.md that explains each. It reads no Runtime.
check [<dir>...]--fix, --explain, --jsonCompiles each agent manifest (default .) offline, with the compiler the Runtime runs, and prints every problem by file, line, and column. --fix makes the edits that fixable problems name; --explain prints each engine feature of the agent and its subagents, on or off, and the tools and text it adds; --json prints the result for CI. It reads no Runtime.
apply [<dir>...]--dry-run, --if-match <n>, --agent [<id>], --watch, --saveSaves the agent manifest each folder describes (default .). See Save agent manifests.
chat <agent>--session <id>, --message <text>, --owner <user>Sends --message, or each line of standard input, to Agent <agent> and prints the reply as it comes. A new session is opened unless --session names one, which is created with that id when it does not exist.
migrate folder <dir>--out <dir>Rewrites a three-file folder (agent.manifest.json) as AGENT.md and its files, with its skills, and says whether it compiles to the same manifest. It reads no Runtime.
migrate folder <dir> --to 2Makes an agent manifest's files agent-folder/2 in place, writing the engine: declarations that keep the text and tools it gets today, after checking that no agent's system text or tool list changes. It reads no Runtime.
migrate folder <dir> --to 3Makes an agent-folder/2 folder agent-folder/3 in place, writing the Runtime's plugins that keep the tools and features each agent gets today, after checking that none changes. See The Runtime's plugins. It reads no Runtime.
export --folder <name>[@<n>]--out <dir>Writes a saved manifest's latest version, or version n, out as files, for one saved as JSON or from TypeScript.
pull <name>[@<n>]--out <dir>Writes the folder a version was saved from, byte for byte.
migrate v8 <agent.json>--runtime <environment-manifest.json>, --name <name>, --out <dir>Writes the three manifest files from an agent manifest v7 (or a workflow manifest v3) and the environment manifest v1 its sessions ran in. It reads no Runtime.
GroupActs throughCredential
agent-manifestsRuntime APIAn application key, or NYLORUN_TOKEN
agentsRuntime APIAn application key, or NYLORUN_TOKEN
sessionsRuntime APIAn application key, or NYLORUN_TOKEN
artifactsRuntime APIAn application key, or NYLORUN_TOKEN
tenantManagement APIA management key
localDocker and kubectl on this machineNone

Which installation

A command acts on one local Tenant, chosen in this order:

  1. --tenant <name>
  2. NYLORUN_TENANT
  3. The Tenant in the project link (.nylorun/link.json, found from the working directory upwards)
  4. For start in a project: the directory's name
  5. Outside a project: default

In a project without a link, every command but start exits 2 and lists the Tenants on this machine. On a local Tenant, the API commands use the project's keys, or keys named cli and cli-management that they create once. They never start a stopped Tenant (exit 3).

Without --tenant, these variables point the API commands at a remote installation:

VariablesActs on
NYLORUN_RUNTIME_URL and NYLORUN_MANAGEMENT_KEYstatus, doctor sandbox, and tenant
NYLORUN_RUNTIME_URL and NYLORUN_SERVER_KEYapply, chat, export, pull, agent-manifests, agents, sessions, and artifacts
NYLORUN_RUNTIME_URL and NYLORUN_TOKENThe same, with your identity provider's token in place of the key: the person it names, with the scopes it grants.

NYLORUN_RUNTIME_URL alone is an app's setting and names nothing.

The Runtime API commands act for a person with --as <subject>, as client.as(subject) does: the Runtime limits the command to that person's own sessions and to the scopes given. --scope <scope> repeats; the default is sessions:own, and the scopes are agents:read, agents:write, sessions:own, and agent-manifests:write.

Input and exit codes

A usage error exits 2 before any request is sent. A secret or an API key never comes from the command line: it is read from standard input or a file, so it stays out of shell history and the process list.

CodeMeaning
0Success.
1Failure, including a missing Docker or Compose v2, an agent manifest that does not compile, or an apply whose plan has errors.
2Usage error, including no Tenant selected, or a moved or removed command.
3The Runtime is not answering, or the local Tenant is not running or does not exist.
4The Runtime port is held by another Host.
5This nylorun pins an older Runtime than the Tenant last ran.
7The containers, the Tenant, or Studio did not become ready.
130 / 143SIGINT / SIGTERM.

Save agent manifests

npx nylorun init agents/triage --example
npx nylorun check agents/triage
npx nylorun apply agents/triage --dry-run
npx nylorun apply agents/triage --agent
npx nylorun chat triage

An agent manifest holds an AGENT.md, with the agent's name, description, and instructions, and optional runtime.json, mcp.json, tools/, triggers/, skills/, files/, and, for a flow, flow.json and agents/. init <path> writes <path>/AGENT.md for the agent named after the path's last segment; --description sets its description, and --example also writes tools/get_order.json, mcp.json, triggers/new-order.md, and a README.md. The first run in a project also writes .gitattributes, adds .nylorun/credentials.json to .gitignore, and maps the folder's JSON files to their schemas in .vscode/settings.json, each only when it is missing.

check compiles each folder offline and prints every diagnostic as file:line:column severity: message, then ok or the count of errors; it exits 1 when a folder does not compile. A folder without an AGENT.md checks each subfolder that has one. A problem one edit fixes ends (fixable with --fix), and check --fix makes those edits, prints each, and checks again; without --fix, check writes nothing. check --json prints { "folders": [{ path, ok, manifest?, diagnostics }] }, with the same exit codes, for CI. See Diagnostics.

apply takes the same folders. It checks each one first, and sends nothing when one does not compile. It then uploads the folder's files by hash, and the Runtime compiles it again and keeps it as the version's source. --dry-run prints the plan, the compiler's notes, and the compiled manifest; a change to the instructions prints as a text diff. --agent then makes the Agent from the manifest, named after it or <id>, or moves it to the version saved, as agents put does, and prints what it needs a person for, such as an unset required secret. An Agent that tracks the manifest moves to each later version.

apply --watch checks and plans again on each change to the folders, and to the files their runtime.json extends, until interrupted; --save also saves each change. Each save is a new version that moves the Agents tracking the manifest, so use it on a development manifest.

chat <agent> talks to the Agent from the terminal. --owner sets the session's owner (default the Agent's owner, else --as, else $USER). A turn that waits for a person prints how to answer it with nylorun sessions, and a turn that fails exits 1. To send JSON input, such as a flow's, use sessions send <id> --data <json>.

Each manifest is planned first: the version it moves from and to, each change (section, path, op), warnings, and errors. With no errors it is saved under If-Match set to the version planned against (0 for a new manifest), its files uploaded first. Any error, a gap between the sections included, saves nothing and exits 1.

FlagEffect
--dry-runPrints the plans, and the manifest compiled from each folder, and saves nothing.
--if-match <n>Saves only over version n. One folder only.
--agent [<id>]Then makes Agent <id> (default the manifest's name) from the manifest, or moves it to the version saved. <id> takes one folder. Write --agent=<id> when <id> names a directory.
--watchChecks and plans again on each change, until interrupted.
--saveWith --watch, also saves each change.

A folder that holds agent.manifest.json instead of AGENT.md is the deprecated three-file format: apply reads its harness.manifest.json and runtime.manifest.json by paths relative to the file that names them, uploads the skills', plugins', and runtime files they name, and saves the composed manifest, with no source kept. It warns on each such folder:

triage: warning: agent.manifest.json (the three-file format) is deprecated: nylorun migrate folder agents/triage writes it as an agent folder (AGENT.md), which saves the same manifest. It is still read until a later release.

A folder with neither applies each subfolder that has either, so npx nylorun apply agents/ saves every agent, of both formats.

migrate folder <dir> rewrites the three-file folder as AGENT.md and its files into --out (default ./<name>), with its skills, and moves each SKILL.md's secrets to AGENT.md. export --folder <name>[@<n>] writes the folder of a saved manifest, and pull <name>[@<n>] the folder a version was saved from. None writes over a file with other bytes. See Convert.

migrate v8 <agent.json> writes agent.manifest.json, harness.manifest.json, and runtime.manifest.json into --out (default .) and overwrites no file. --runtime takes the environment manifest v1 the agent's sessions ran in: its fields become the runtime section. Without it, the kind is pod when the agent required one, virtual when it has skills or required a sandbox, and none otherwise. --name sets the manifest's name (default: the agent's id). Then convert them with nylorun migrate folder <dir>, or save them with nylorun apply <dir>.

agent-manifests

The Runtime API. Save a manifest with apply. Every command takes --tenant <name> and --as <subject>.

CommandFlagsNotes
list--jsonEach manifest's latest version: its runtime kind, harness, and the Agents made from it.
get <name>[@<n>]A manifest's latest version, or version n, as JSON.
versions <name>Its versions, with the Agents made from each.
rm <name>Deletes every version. Refused while an Agent was made from one.

agents

The Runtime API: the Tenant's Agents, each made from a saved agent manifest. Every command takes --tenant <name> and --as <subject>.

CommandFlagsNotes
list--state <state>, --manifest <name>, --label <key=value>, --outdated, --limit <n>, --cursor <c>, --jsonId, manifest version (tracking or pinned, and the version it moves to), kind, state, sessions, owner, and labels. States: working, sleeping, input_required, failing, paused. --outdated (alias --behind) lists the Agents moving to another version.
get <id>The Agent, as JSON.
put <id> --manifest <name[@n]>--owner <user>, --label <key=value>, --dry-runCreates the Agent from a saved manifest: name tracks its latest version, name@n pins version n. On an existing Agent, changes its owner and labels and moves it to that version. --dry-run prints what the move would do.
pin <id> <version>Moves it to that version of its manifest and keeps it there.
track <id>Moves it to its manifest's latest version, and again with each save.
restart <id>--retryApplies the version it waits to start on now: a running pod stops, and the next turn starts it. --retry tries the version that failed to start again.
fork <id> <new-id>--manifest <name[@n]>, --disk latest|<n>|empty, --secrets copy|none, --label <key=value>, --dry-runA new Agent from a manifest version (default: its manifest's latest), with this one's disk and the secrets it declares. --dry-run prints the plan. See Fork.
stop <id>Stops its sandbox.
reset <id>Resets its disk.
rm <id>Deletes it with its sandbox, files, secrets, and memories. Refused while a turn runs on it.
snapshot <id>Keeps a version of its disk.
versions <id>Its disk versions.
restore <id> <version|latest>Puts a disk version back.
delete-version <id> <version>Deletes a disk version.
events <id>--from <n>, --types <type,…>, --since <time>, --until <time>Its events as JSON: lifecycle, and agent.egress for every connection, naming the turns running.
secrets list <id>Names, types, and bindings; never a value.
secrets get <id> <name>One secret, without its value.
secrets put <id> <name>--via <url>, --tool-prefix <prefix>The value of a slot its manifest declares, from standard input, with where its requests go. See Route a slot.
secrets put <id> <name> <secret.json|->Creates or replaces a typed secret, such as {"type": "bearer", …}, from a file or standard input.
secrets rm <id> <name>Deletes a secret.
triggers list <id>Its triggers: status, next and last run, as JSON.
triggers put <id> <trigger>--schedule <cron>, --timezone <zone>, --at <time>, --webhook, --session <id|new>, --prompt <text>, --title <text>, or a JSON filePuts a trigger of the Agent's own (origin api), or replaces it; the same trigger again changes nothing. See A client's triggers.
triggers rm <id> <trigger>Removes a trigger an agent made or a client put. A manifest's trigger can't be removed this way.
triggers pause <id> <trigger>Pauses a trigger.
triggers resume <id> <trigger>Resumes a paused trigger.
triggers run <id> <trigger>--idempotency-key <k>Runs it once now, as if it fired.
triggers new-secret <id> <trigger>A webhook trigger's new signing secret (whsec_…), shown once with its URL. The previous secret stops working.
runs <id>--trigger <id>, --limit <n>, --cursor <c>What each fire came to (started, queued, skipped, failed), newest first, as JSON.
memories ls <id>--prefix <folder>, --limit <n>, --cursor <c>Its memories by path, as JSON.
memories cat <id> <path>One memory's content.
memories put <id> <path> <file|->Writes a memory whole, from a file or standard input, such as to seed one.
memories rm <id> <path>Deletes a memory, or every memory in a folder.
memories versions <id>--path <path>, --limit <n>, --cursor <c>Every change to its memories, newest first, as JSON.

sessions

The Runtime API. Every command takes --tenant <name> and --as <subject>.

CommandFlagsNotes
list--agent <id>, --manifest <name>, --status <status>, --owner <user>, --limit <n>, --cursor <c>, --json--agent: the sessions on that Agent. --manifest: the sessions on the Agents made from that agent manifest.
create --agent <id> --owner <user>--id <id>, --info <json>Opens a session on Agent <id> and prints its id.
get <id>Status, active turn, and what it waits for.
send <id> [<text>]--data <json>, --file <artifact>[@<version>], --idempotency-key <k>, --reject-if-busyStarts a turn with text, JSON data, or files that are artifacts. --file repeats. While a turn runs or waits, the message waits in the session's queue and starts the next turn; --reject-if-busy refuses it instead (409 session_busy).
approve <id> <interaction>--denyApproves, or denies, a held tool call.
respond <id> <interaction> <json>Answers a question the agent asked.
cancel <id>--reason <text>Cancels the running turn. Queued messages stay queued.
watch <id>--follow, --cursor <c>The session's events as JSON lines until interrupted. --follow adds the sessions of agents it uses as tools.
history <id>--agent <a>, --limit <n>, --cursor <c>
usage <id>--turn <id>Tokens and cost.
calls <id>--tool, --turn <id>, --limit <n>, --cursor <c>Model calls, or tool calls with --tool.
trace <id> <turn>The turn's spans as OTLP JSON.
exec <id> -- <command>...--timeout <s>Runs a bash command in the session's sandbox and exits with its exit code.

artifacts

The Runtime API. Every command takes --tenant <name> and --as <subject>.

CommandFlagsNotes
list--session <id>, --label <key=value>, --limit <n>, --cursor <c>, --jsonNewest first: id, kind, name, latest version, owner.
get <id>The artifact and its versions, as JSON.
upload <file|->--name <name>, --content-type <type>, --session <id>, --label <key=value>, --to <id>Uploads a file as a new artifact, or as the next version of --to. Prints its id.
download <id>--version <n>, --output <file>The bytes, to a file or standard output.
tree <id>--version <n>A folder artifact's files.
diff <id>--version <n>, --from <n>What changed between two versions of a folder.
zip <id> --output <file>--version <n>A folder artifact as a zip.
link <id>--version <n>, --expires-in <s>, --file <path>A link anyone holding it may download from.
rm <id>Deletes the artifact and its versions.

tenant

The Management API, with a management key. Every command takes --tenant <name>.

CommandFlagsNotes
reset--sessions, --agents, --all, --yesClears the Tenant's sessions (the default), its Agents with their secrets, disks, and sessions, or all its data. The Tenant, its keys, and the project link stay. Pass one scope at most (exit 2). --all asks first; pass --yes when not in a terminal.
keys put <id>--managementCreates or rotates application key <id> (a management key with --management, on the Runtime's machine only). Prints it once; a rotated key stops working.
keys list--jsonIds, roles, and issue times; never the keys.
keys rm <id>The key stops working at once.
signing-keys listThe Tenant's token signing keys.
signing-keys rotate--force--force ends outstanding links and run tokens.
signing-keys revoke <kid>Revokes a previous or standby key.
models setSigns in to a model provider and sets the Tenant's model. See Models.
models set --provider <id> --model <id>--base-url <url>The same with an API key piped in on standard input, for CI. A terminal on standard input is refused (exit 2).
models select <provider>--model <id>Uses a provider that already has a credential (default: its model).
models get--jsonThe Tenant's model.
models providers--jsonThe providers with a credential, and the one in use.
models catalog--jsonEvery provider and model the Runtime knows.
usage--scope tenant|manifest|turn, --id <id>, --period day|month|totalWhat model calls used and cost, as JSON. --scope manifest --id <name>: the Agents made from that agent manifest.
calls--after <cursor>, --limit <n>Every recorded model call, one JSON line each, until caught up. Prints the cursor to resume from on stderr. A page holds at most the Runtime limit exportPageItems.
budgets [get | put <file|->]The model spend caps: scope tenant, manifest (with id an agent manifest's name), or turn. put replaces them all. See Usage and budgets.
tool-policy [get | put <file|->]The tool policy. put replaces every rule.
settings sandbox [get | put <file|->]The Agents' compute settings: limits, stop grace, placement, engine images. get prints config (as set) and effective (in force); put takes what get printed and puts back its config.
settings artifacts [get | put <file|->]The artifact store's limits.
settings skills [get | put <file|->]The most bytes one skill may hold.
settings limits [get | put <file|->]The Runtime limits: sizes, counts, and deadlines.
coverage --agent <id>--jsonWhich secret of Agent <id> each MCP server and HTTP tool of its manifest would get. See Coverage.
mcp inspect <url>--server <name>, --agent <id>, --sse, --jsonLists a remote MCP server's tools, sending Agent <id>'s credential for <url> (none without --agent). Calls no tool. See Preview an MCP server.

budgets, tool-policy, and settings are documents. get (the default) prints one as JSON. put <file>, or put - for standard input, checks the body against its schema before sending it and replaces the whole document. What budgets, tool-policy, settings artifacts, settings skills, and settings limits print, put takes back, so they can live in a repository and be applied from CI:

npx nylorun tenant tool-policy get > tool-policy.json
npx nylorun tenant tool-policy put tool-policy.json

settings limits get prints every limit in force, defaults included, so putting that file back fixes today's defaults as the Tenant's own values. To keep the defaults, keep only the limits you change in the file.

On a remote installation, tenant keys manages application keys only and refuses --management.

local

The Tenants on this machine, through Docker Compose and kubectl.

CommandFlagsNotes
ls--jsonThe Tenants on this machine, with their state, memory in use, and URLs.
delete <tenant> --yesRemoves a Tenant: containers, volumes, files, and the vault key. --yes is required.
logs [service]--tenant <name>, -f, --tail <n>Container logs. Services: postgres, restate, s2-lite, rustfs, gateway, runtime, harness, studio, and sandboxes.
reset--tenant <name>, --yesDeletes the Tenant's volumes, Tenant directory, and vault key. The next start creates it anew.
sandbox enable --context <name>--tenant <name>, --backend agent-sandbox|substrate, --host-address <ip>, --bind-address <ip>, --no-pullRuns the Tenant's sandboxes as pods on that Kubernetes context (preview). See Kubernetes. --backend substrate runs them as Agent Substrate v1.0 actors instead, on Kubernetes 1.37 with Substrate installed (preview). See Agent Substrate.
sandbox disable--tenant <name>, --delete-namespaceStops the sandboxes service and forgets the cluster.
sandbox status--tenant <name>, --jsonThe cluster and the sandboxes service's readiness; on Agent Substrate, also the atespace and the edge.
telemetry [status|enable|disable]Studio's anonymous page-view analytics. Off with NYLORUN_TELEMETRY_DISABLED=1, DO_NOT_TRACK=1, or in CI.

local reset and tenant reset differ. See Local Tenants.

The CLI's own waits

How long nylorun waits for the local stack, how often it polls, the healthchecks it writes into the Compose file, and the waits of nylorun local sandbox are its tuning: one JSON object in NYLORUN_CLI_TUNING, in the shell that runs nylorun. A setting left out keeps its default. An unknown key, or a value that is not a positive whole number, stops the command (exit 2).

NYLORUN_CLI_TUNING='{"stack":{"healthWaitMs":120000}}' npx nylorun start
GroupSettings (defaults)
stackprobeTimeoutMs (3 s), healthWaitMs and tenantWaitMs (60 s), pollMs (500 ms), operateEveryPolls (10), composeWaitSeconds (300) and studioWaitSeconds (120) for docker compose up --wait-timeout, loginWaitMs (15 s), loginRequestMs (5 s), keyCheckMs (10 s), the key file's lockWaitMs (30 s), lockStaleMs (60 s), and lockPollMs (100 ms), portAttempts (20), seedRequestMs (15 s)
composewalKeepSizeMb (4096, Postgres's max_slot_wal_keep_size), and healthchecks.<service> as { intervalSeconds, timeoutSeconds, retries }: postgres, restate, rustfs, and studio (2, 5, 30), gateway and runtime (2, 5, 60), harness (2, 5, 90), sandboxes (5, 5, 24)
sandboxkubectlTimeoutMs (300 s), kubectlMarginMs (20 s), manifestDownloadMs (60 s), crdWaitSeconds (60), rolloutWaitSeconds (180), tokenWaitMs (60 s), pollMs (1 s), composeWaitSeconds (120), infoRequestMs (5 s), probeWgetSeconds (3), probeExecMs (20 s), probeReadySeconds (180), policyTimeoutMs (30 s), prepullMs (300 s), prepullPollMs (2 s), the probe pods' probeGraceSeconds (1; 0 allowed), probeCpu (200m), and probeMemory (64Mi)

On a slow machine, raise stack.healthWaitMs and the harness healthcheck's retries first. The Runtime's own timings are its Host tuning (NYLORUN_TUNING), and the sizes and deadlines a Tenant is held to are its Runtime limits.

Moved and removed commands

A command that moved exits 2 naming its replacement:

BeforeNow
nylorun environments …, every subcommandnylorun agents …. Save manifests with nylorun apply; convert an environment manifest with nylorun migrate v8.
nylorun agents save <manifest.json>nylorun apply <dir>, then nylorun agents put <id> --manifest <name>
--environment <id> on sessions create, sessions list, tenant coverage, and tenant mcp inspect--agent <id>
nylorun tenant reset --environmentsnylorun tenant reset --agents
nylorun tenant usage --scope agentnylorun tenant usage --scope manifest, with --id an agent manifest's name
--scope environments:write--scope agents:write, which creates, changes, and deletes Agents and writes their secrets. Saving agent manifests is --scope agent-manifests:write.
nylorun key put|list|rmnylorun tenant keys put|list|rm
nylorun mcp inspect <url>nylorun tenant mcp inspect <url>
nylorun configurenylorun tenant models set
nylorun sandbox ls|rmnylorun agents list|rm
nylorun sandbox enable|disable|statusnylorun local sandbox …
nylorun logs, ls, delete, telemetrynylorun local …
nylorun resetnylorun local reset (the local Tenant's volumes) or nylorun tenant reset (sessions or Agents, through the Management API)

nylorun dev, nylorun serve, nylorun runtime, nylorun env, and nylorun tenant mcp connect were removed; each exits 2 and says what to do instead.

nylo, the second command of the nylorun package, is part of nylorun. It still ships, and a later release removes it. Each nylo command exits 2 naming its replacement:

nylonylorun
nylo statusnylorun status
nylo reset […]nylorun tenant reset [--sessions|--agents|--all]
nylo access signing-keys …nylorun tenant signing-keys …
nylo configurenylorun tenant models set
nylo doctor sandboxnylorun doctor sandbox
nylo envNone

nylo env has no replacement: the SDKs read the project link themselves, so the project's npm run dev and npm start need no Nylorun tool. For a process outside the project, give it a key of its own (nylorun tenant keys put <id>) and the Runtime URL that nylorun status prints.

@nylorun/cli is nylorun under another name: npx @nylorun/cli <command> runs nylorun <command>, with the same help, output, and exit codes. It has no nylo. Install one of the two globally, not both: each provides nylorun.

npm create @nylorun/agent

npm create @nylorun/agent@beta <dir> [--yes]

Writes the starter project and installs dependencies. It starts nothing.

Next step

On this page