@deepraxis/cli weekly downloads statistics

<div align="center">

Praxis CLI

Models, agents, sandboxes, search and data from one terminal, with one account.

The official command line for 知行云 Praxis, the cloud where AI agents are built, run and hosted.

Website · Documentation · API reference · Console

</div>

---

npm install -g @deepraxis/cli
praxis login
praxis api model chat completion create --model deepseek-flash \
  --data '{"messages":[{"role":"user","content":"Hello!"}]}'

Every operation of the Praxis API is a command under praxis api: about 350 of them across models, agents, sandboxes, cloud browsers, web search, databases, knowledge bases, files, media, automation and hosted applications. The command tree is generated from the same contract as the API and the SDK, so a new API capability reaches the CLI on the day it ships.

Build and deploy an application

praxis app create my-app         # standard template, dependencies installed
cd my-app && praxis app deploy   # build, publish, create declared secrets, health check, print the URL
praxis app rollback              # deploy the previous version again

praxis.jsonc at the project root names the app and may declare "secrets": {"SESSION_SECRET": "generate"} (created once, kept by every later deploy) and "healthcheck": "/api/health" (a failing check fails the deploy). In computers and sandboxes the CLI signs in with PRAXIS_API_TOKEN; PRAXIS_API_KEY takes precedence when both are set.

Why Praxis CLI

• One account, every capability. Chat with models from many providers, start a sandbox, search the web and query a database without juggling vendor keys.
• Made for people and for AI agents. Readable tables in your terminal, compact text when an AI coding agent runs it, and clean JSON in scripts. The CLI picks the mode by itself.
• Signs in through the browser. praxis login approves the session in your browser and keeps the credential in the system keychain, never in a dotfile when a keychain exists.
• Predictable grammar. praxis api <domain> <resource> <verb> [id], the same shape everywhere, with --help on every level.

Install

Requires Node.js 20 or newer.

npm install -g @deepraxis/cli     # or: pnpm add -g / bun add -g
praxis --version

Run it once without installing:

npx @deepraxis/cli model list

Sign in

praxis login    # approve in the browser with the code shown
praxis whoami   # account, organization, credential storage

You need a Praxis account; sign up at deepraxis.com.

Belong to several organizations? Every call runs in the active one:

praxis org list
praxis org switch <organization_id>
praxis api model list --org <organization_id>   # just this call

For CI and other headless environments, create an API key in the Console and set it instead of logging in. PRAXIS_API_KEY always takes precedence over a stored login:

export PRAXIS_API_KEY=<your-api-key>

A tour

Models

praxis api model list
praxis api model chat completion create --model deepseek-flash \
  --data '{"messages":[{"role":"user","content":"A haiku"}]}'
praxis api model embedding create --model <embedding_model> \
  --data '{"input":"hello world"}'
praxis api model image generation create --model <image_model> \
  --prompt "a lighthouse at dawn"

Agents

Define an agent once, then talk to it in sessions:

praxis api agent create --data '{
  "name": "helper",
  "model": {"id": "deepseek-flash"},
  "tools": [{"type": "agent_toolset_20260401"}]
}'
praxis api agent session create --data '{"agent": "<agent_id>"}'
praxis api agent session event send <session_id> --data '{
  "events": [{
    "type": "user.message",
    "content": [{"type": "text", "text": "What can you do?"}]
  }]
}'
praxis api agent session event stream --session <session_id>

Sandboxes

praxis api sandbox create --template-id base
praxis api sandbox command run <sandbox_id> \
  --cmd "python3 -c 'print(6 * 7)'"
praxis api sandbox delete <sandbox_id> --yes

Search, data and automation

praxis api search create --query "AI agent frameworks" --max-results 5
praxis api database create --name notes
praxis api automation run create --data '{
  "source": "export default async (i) => ({ answer: i.n + 1 })",
  "language": "ts", "input": {"n": 41}, "wait": 30
}'

Browse everything with praxis --help, praxis api --help, praxis api <domain> --help, or the full command reference that ships in the package.

How commands work

There are two layers. Top-level commands are workflows and session commands (login, logout, whoami, org, skill). Every API operation lives under praxis api, so the two never collide.

praxis api  sandbox  command  run  sbx_123  --cmd "ls"  --json
            domain   resource verb target   options     global
```…