Agent-friendly CLI for managing & debugging Upstash resources from your terminal. Docs.
Requires Node.js 20 or newer.
npm i -g @upstash/cliFor agents, pair the CLI with the Upstash Skill, it bundles Upstash docs alongside docs for the upstash CLI.
npx skills add upstash/skillsIf you are an AI agent (or just want to try things out) and need a Redis DB to use or experiment with, you don't need an Upstash account:
upstash start-redisThis prints markdown with credentials and a quickstart. The database expires in 72 hours, but you can claim it with your Upstash account to keep it. Pass --id <database-id> to re-fetch the credentials of a database you created earlier. Every command that touches your Upstash account needs authentication — only start-redis and the login/logout credential helpers do not.
If you are an agent, identify yourself with --user-agent on every start-redis call, so we can see which agents are creating databases:
upstash start-redis --user-agent claude-code # or cursor, codex, opencode, ...Grab a Developer API key from the Upstash Console, then save it once per machine:
upstash loginOr set UPSTASH_EMAIL and UPSTASH_API_KEY in your shell or a .env file. See the auth docs for env files, per-command flags, and precedence rules.
Every command that returns account data outputs JSON, so you can pipe to jq. The exceptions are start-redis, which prints markdown, and login/logout, which print a plain-text confirmation. Use --dry-run to preview destructive commands.
# Redis
upstash start-redis # free temporary DB, no account needed
upstash redis list
upstash redis create --name my-db --region us-east-1
upstash redis exec --db-url $URL --db-token $TOKEN GET key
# Vector
upstash vector list
upstash vector create --name my-index --region us-east-1 --similarity-function COSINE --dimension-count 1536
# Search
upstash search list
upstash search create --name my-search --region us-central1 --type DENSE
# QStash
upstash qstash list
upstash qstash stats --qstash-id $QSTASH_ID --period 7d
# Blob
upstash blob create --name my-bucket --visibility private
upstash blob ls
upstash blob ls my-bucket
upstash blob cp ./assets blob://my-bucket/assets -r
upstash blob sync ./site blob://my-bucket/site -d
upstash blob credentials my-bucket
# Team
upstash team list
upstash team add-member --team-id $TEAM_ID --member-email you@example.com --role devRun upstash --help (or --help on any subcommand) to discover everything else, and check the full docs for the complete catalog. upstash blob credentials returns temporary S3 credentials for use with AWS CLI, rclone, or an S3 SDK.
The object commands mirror aws s3, with blob://<bucket>/<key> in place of
s3://. <bucket> is a bucket name or id.
upstash blob ls # buckets
upstash blob ls my-bucket/images/ # one level; -r for all
upstash blob cp ./photo.png blob://my-bucket/images/
upstash blob cp ./assets blob://my-bucket/assets -r
upstash blob cp blob://my-bucket/images ./images -r --exclude "*.tmp"
upstash blob cp blob://my-bucket/config.json - | jq .
upstash blob mv blob://my-bucket/a.txt blob://other-bucket/a.txt
upstash blob sync ./site blob://my-bucket/site -d
upstash blob rm my-bucket/tmp -r -n
upstash blob presign my-bucket/report.pdf --expires-in 600
upstash blob mb blob://new-bucket
upstash blob rb new-bucket -fcp, mv and sync need blob:// to tell bucket paths from local ones. The
commands that only take bucket paths (ls, rm, presign, mb, rb) accept
my-bucket/key without it, as do get, delete and credentials, which take a
bucket name or id.
Flags follow aws s3: -r/--recursive, --exclude/--include (applied in
order, last match wins), -n/--dryrun, -d/--delete, --size-only,
--exact-timestamps, --content-type, --cache-control, --metadata,
--expected-size, --concurrency and -q/--quiet. Copies between buckets reset
Cache-Control to the default unless --cache-control is given. Local symbolic
links are followed unless --no-follow-symlinks is given. Nothing is deleted
through a link to a directory: mv refuses such files and sync -d keeps them.
Progress goes to stderr and a JSON summary to stdout. Transfers retry transient failures, keep going past a failed file, and exit unsuccessfully at the end. Large files use multipart uploads, and the Blob SDK refreshes temporary S3 credentials throughout, even between parts of one file.
Bucket names need an Upstash login. A Blob bucket token (--token, or
UPSTASH_BLOB_TOKEN in the environment or .env) works without one, but only for
its own bucket, addressed by id. A token is never used for a bucket it wasn't
issued for.
upstash blob cp ./assets blob://$BUCKET_ID/assets -r --token "$BLOB_TOKEN"
upstash --env-path ./uploads.env blob sync ./assets blob://$BUCKET_ID/assets
upstash blob credentials --token "$BLOB_TOKEN"--token and UPSTASH_BLOB_TOKEN are each used only for their own bucket, so they
can point at different buckets. Exported environment variables take precedence
over values loaded from .env or --env-path.
The CLI identifies itself to the Upstash API on each request, so we can see which clients our endpoints are serving. It sends three headers and nothing else:
| Header | Example |
|---|---|
Upstash-Telemetry-Sdk |
@upstash/cli@1.2.0 |
Upstash-Telemetry-Runtime |
node@22.14.0 |
Upstash-Telemetry-Platform |
darwin |
That is the CLI version, the JS runtime, and the OS platform. No command arguments, credentials, resource names, or file paths are collected.
To turn it off:
upstash telemetry disable # saved to your config file
upstash telemetry status # check the current setting
upstash telemetry enable # turn it back onOr set the environment variable every Upstash SDK honors, which also works from
a .env file and takes precedence over the saved setting:
export UPSTASH_DISABLE_TELEMETRY=1Disabling telemetry never affects what the CLI can do. upstash logout keeps
the setting, so signing out does not quietly turn it back on.
npm install
npm run build
node dist/cli.js --help # try your build
npm link # or expose it as `upstash` globallyOpen an issue, send a PR, or join us on Discord.