Skip to main content
This page covers every command available in @praxa/cli@0.3.0. Each section includes a description of what the command does, the full flag set, which environment variables it reads, and a runnable example. Gateway commands (doctor and mission), version, and memory planning write JSON to stdout. praxa init requires --json for JSON output, help output is text, and errors are written to stderr with exit code 1 on failure. For doctor and mission commands, the gateway URL resolves from --base-url, then PRAXA_BASE_URL (or legacy AURA_BASE_URL), then .praxa/config.json in the current working directory. PRAXA_ACCESS_TOKEN remains an environment-only secret.

praxa version

Prints the CLI version, the embedded OpenAPI spec version, the wire contract version, and a SHA-256 fingerprint of the OpenAPI source. This command makes no network connection and requires no configuration, making it a reliable smoke test after installation. Syntax
Flags This command accepts no flags. Environment variables None required. Example output
Run praxa version immediately after installation to confirm the correct build is on your PATH and to capture the exact OpenAPI version for support requests.

praxa init

Configures one or more project-scoped agent clients without writing an access token into project files. Syntax
Use --target repeatedly to select codex, claude, cursor, vscode, or env; omit it to configure all targets. --dry-run reports intended changes without writing files, and --force replaces a conflicting managed entry.

praxa memory source add

Previews or writes one read-only memory-provider entry in .praxa/memory.json. The command accepts no provider credential and makes no provider or gateway request. Syntax
<provider> must be mem0, zep, graphiti, langgraph, letta, or openai_agents. --mode federated is required. Use --dry-run to return the exact local file plan without writing.
The result identifies the project directory, .praxa/memory.json action, provider, read-only access, and local-only execution boundary. It never reports a provider connection or sync.

praxa memory sync plan

Reads local source metadata and returns a bounded, non-executable plan. It cannot start a sync, mirror, import, cutover, provider write, or migration. Syntax
--dry-run is required. The returned executable field is always false and operations is empty in 0.3.0.

praxa doctor

Makes an authenticated read-only request to your gateway to verify that the URL is reachable, the TLS certificate is valid, and the access token is accepted. It fetches reference coverage data and returns it alongside an ok: true field on success. Syntax
Flags Environment variables Example

praxa mission submit

Records natural-language intent for deterministic compilation. A successful response is durable admission, not proof of execution or a provider outcome. Syntax

praxa mission get

Retrieves an authoritative mission projection for the given run ID. Use this command to poll mission status, inspect the current state, or retrieve output after a mission completes. Syntax
Flags Environment variables Example
You can pipe the output to jq to extract a specific field:

praxa mission create

Submits a new mission to the Praxa Integration Gateway by reading a JSON file from disk. The response contains the assigned run ID and the initial mission state. Syntax
Flags Environment variables Mission file format Your --input file must be valid JSON with at minimum a goalSpec object and a resourceBudget object:
Example
Store the returned run ID for subsequent get or cancel calls:

praxa mission cancel

Requests durable cancellation of a running mission. The reason string is recorded in the mission audit log. Create one stable idempotency key for this logical cancellation and reuse it if a network failure requires a retry. Syntax
Flags Environment variables Example

Gateway commands (doctor and mission), version, and memory planning write pretty-printed JSON to stdout, making them easy to pipe to jq. Pass --json to praxa init when you need machine-readable output; help remains human-readable text. Errors are written to stderr and the process exits with code 1.
Idempotency keys must contain 16–128 supported characters and match ^[A-Za-z0-9][A-Za-z0-9._:-]{15,127}$. Create one key per logical mutation and reuse it for every retry of that same mutation. A value from $(uuidgen) is valid, but UUIDs are not required. Never reuse a key for a different mutation because the gateway binds the key to the original request.
Last modified on August 14, 2026