CLI command and API reference

CLI command reference

For released @evolinkai/cli 0.8.1. Use evolink —help or evolink COMMAND —help —json for your installed version. Business commands support —json. Replace MODEL_ID, QUOTE_ID, TASK_ID and UPLOAD_ID with actual returned values.

Authentication, diagnostics and skills

See installation.

Command	Behavior
evolink —version	Version
evolink setup —agent codex —json	Prerequisites, login, skill sync and free connection verification
evolink auth login [—no-browser] [—timeout SECONDS]	OAuth, 30–900 seconds, default 180
evolink auth status —json	Local login state, not network verification
evolink auth logout	Revoke the current CLI session and sign out
evolink doctor —agent codex —json	Environment, storage, account, models and skills
evolink skills install [—agent NAME] [—replace-modified]	Install/sync; preserve edits by default
evolink skills status [—agent NAME] —json	Missing, current, outdated, modified or conflicting skills

Agent values: all/codex/claude-code/cursor/gemini/opencode/copilot/openclaw/hermes; default all. The assistant must still verify skill discovery.

Models, inputs and documentation

evolink models search --type image --query seedream --limit 5 --page 1 --json
evolink models show MODEL_ID --json
evolink models schema MODEL_ID --json
evolink models recommend --type video --query seedance --references image --limit 3 --json
evolink docs search --type video --query "first frame" --limit 5 --json

Command	Options and output
models search	type image/video/audio/all, default all; limit 1–50, default 20; page 1–100000; query up to 100 characters
models show MODEL_ID	Canonical ID, aliases, parameters, prices, references, documentation provenance and schema
models schema MODEL_ID	Versioned input/submission schema; explicitly fails when unavailable
models recommend	Required type image/video/audio; references comma-separated image/video/audio; limit 1–10
docs search	Required query up to 100 characters; type as search; limit 1–20, default 5

Docs search uses the bundled official model-reference index, not a live website crawl. Recommendation is based on documented inputs, keywords and editorial platform preferences, not a popularity leaderboard. Unit prices are not task quotes.

Quotes and generation

Create input.json using the model’s real parameters, then obtain a free quote:

evolink estimate --model MODEL_ID --input-file input.json --max-cost-usd 1 --json

Option	Meaning
—model	Required canonical ID or supported alias
—input-file FILE / —input JSON	JSON object, choose one; omission starts with an empty object
—prompt TEXT	Prompt shortcut; cannot conflict with JSON prompt
—max-cost-usd NUMBER	Positive submission estimate cap, at most USD10000
—media-seconds NUMBER	Positive pricing hint up to 3600 seconds, not a model input

A successful quote returns quote_id, lasts about 15 minutes, and binds login, backend, model and exact inputs. Failed/cap-refused estimates are not usable quotes. Show specifications, range and warnings, wait for explicit approval, then select the matching command:

evolink generate image --quote QUOTE_ID --confirm --wait --timeout 1800 --json
evolink generate video --quote QUOTE_ID --confirm --json
evolink generate audio --quote QUOTE_ID --confirm --json

These are separate examples; one quote covers one type/task. Generation accepts no replacement model/input. Requote and approve changes. —confirm declares approval was obtained; it cannot ask the user by itself. —wait waits for the existing task. Timeout 1–86400 seconds, default 1800. Timeout/Ctrl-C does not cancel the server task. Complete estimates default to their quoted maximum for submission checks. See budget limitations.

Tasks, batches and recovery

evolink tasks get TASK_ID --json
evolink tasks wait TASK_ID --timeout 1800 --json
evolink tasks batch --ids TASK_ID_ONE,TASK_ID_TWO --json
evolink tasks list --type video --status processing --since 2h --limit 20 --page 1 --json
evolink tasks resume --quote QUOTE_ID --json

List option	Contract
—status	processing/completed/failed/cancelled; processing includes queued; omit for all
—type	image/video/audio; omit for all media
—model	Exact canonical model ID
—since / —until	Creation time: ISO8601, Unix seconds, 30m/2h/1d; until inclusive
—page / —limit	Page 1–100000; size 1–50, default 20

Time filtering affects only the selected page. History covers the account, not just this chat. An empty page cannot prove submission was absent; total precedes local time filtering. Batch IDs are 1–50 comma-separated values. With a task ID, use get/wait. For a lost submit response, resume the original quote with its request ID and login. If records are lost, expired or the account changed, inspect history/console before another paid intent. CLI has no cancellation command.

Balance and task usage

evolink balance --json
evolink usage --since 7d --until 1h --type video --max-pages 5 --json

Balance returns account balance and applicable limits. Usage accepts model/type/creation time; default since 30d, max-pages 1–20, default 5, 50 tasks per page. Inspect missing costs, coverage and truncation. Usage is reported completed-task cost from bounded account-wide retained history, not a bill, refund ledger, CLI/MCP-only usage or final budget.

Uploads and original downloads

evolink upload ./reference.mp4 --upload-path references --json
evolink uploads get UPLOAD_ID --json
evolink download TASK_ID --output ./result.mp4 --index 1 --json
evolink download TASK_ID --all --output-dir ./results --json
evolink download TASK_ID --all --output-dir ./results --resume --json

CLI accepts regular files from 1 byte to 95 MiB, saving an upload ID first. uploads get only reads the original receipt. Without compatible backend support, it can report unknown without resending. References typically remain for 72 hours. Download index is 1-based, default 1, range 1–50. Parent directories must exist; files are not overwritten. Each result is at most 1 GiB. Type/content checks reject HTML error pages masquerading as media. Batch mode —all —output-dir DIR supports —template NAME and —resume. Resume uses original batch records and does not trust arbitrary same-name files. More in task delivery.

JSON, exit codes and configuration

—json writes one stdout JSON line with schema_version:1 and ok. Progress goes to stderr. Errors use ok:false/error and nonzero exit; Ctrl-C usually exits130. A task status:failed differs from a query invocation failure. Check command ok, then task state. Local records remain in ~/.evolink-media; credentials use OS secure storage. Do not copy this directory to migrate login or delete quote/recovery records to fix an uncertain submission. Advanced —server selects the OAuth resource/identity binding, default https://mcp.evolink.ai/mcp, without changing REST execution. Keep —api-url/—files-url production defaults; arbitrary external URLs are refused. —token-stdin is a one-command OAuth access token, not a platform API key, and is not for setup. Do not put tokens in command arguments or logs. No generic run, text chat, cancellation, top-up, key management or custom alias-file command is provided. Use the corresponding console/API.

CLI execution path and APIs

Since CLI 0.8.0, media requests use terminal or agent → EvoLink CLI → platform REST API. Input checks, model references, estimation, and result formatting run locally; business execution does not call remote MCP tools/call.

Service	Address and role
Platform API	https://api.evolink.ai: models, media, tasks, and balance
Passport	https://passport.evolink.ai: browser authorization and refresh
Files	https://files-api.evolink.ai: uploads and compatible receipts
OAuth resource	https://mcp.evolink.ai/mcp: identity binding, not the media execution path
Generated originals	HTTPS URLs returned by tasks: direct download without platform tokens

Commands and backend operations

CLI capability	Backend operation or implementation
models search	`GET /v1/models`, combined with aliases, references, and availability
models show/schema, models recommend, docs search	Published bundled references, shared rules, and current account catalog
estimate	`GET /web/api/models/pricing` and client estimation rules
generate image	`POST /v1/images/generations`
generate video	`POST /v1/videos/generations`
generate audio	`POST /v1/audios/generations`
tasks get/wait	`GET /v1/tasks/{task_id}`
tasks batch	`POST /v1/tasks/batch`
tasks list, usage	`GET /v1/tasks`, pagination, and bounded aggregation
balance	`GET /v1/credits`
upload	`POST /v1/files/upload-token`, followed by direct file-service upload
uploads get	Compatible file service `GET /api/v1/files/upload-receipts/{upload_id}`
download	Direct task result download, format checks, and result records

CLI quote_id, —confirm, and recovery files implement a client workflow, not new backend fields. Published CLI 0.8.1 still uses public-default estimation; the existence of new pricing APIs does not mean this version consumes account-specific media quotes. —max-cost-usd is not a final settlement cap.

Requests, recovery, and automation

Persist the quote and request identifier before submitting. REST uses existing Idempotency-Key / X-Evo-Run-Id identifiers. Recover with the original quote after a lost response; do not change identity, inputs, or identifiers to resubmit. Client records alone do not prove uniform backend idempotency across production nodes. Scripts should inspect ok, then task status, then results and cost fields. Store quote_id, task_id, and download manifests instead of relying on the last terminal line. Query errors, failed tasks, and failed downloads require different handling. Do not unconditionally retry paid submissions. CLI and native MCP sessions are managed separately, but account OAuth currently shares the permissions, limits, and pause state of EvoLink MCP (OAuth). Use the separate MCP documentation for native connections.

Start here

CLI

MCP

Shared resources

CLI command and API reference

CLI command reference

Authentication, diagnostics and skills

Models, inputs and documentation

Quotes and generation

Tasks, batches and recovery

Balance and task usage

Uploads and original downloads

JSON, exit codes and configuration

CLI execution path and APIs

Commands and backend operations

Requests, recovery, and automation

​CLI command reference

​Authentication, diagnostics and skills

​Models, inputs and documentation

​Quotes and generation

​Tasks, batches and recovery

​Balance and task usage

​Uploads and original downloads

​JSON, exit codes and configuration

​CLI execution path and APIs

​Commands and backend operations

​Requests, recovery, and automation

CLI command reference

Authentication, diagnostics and skills

Models, inputs and documentation

Quotes and generation

Tasks, batches and recovery

Balance and task usage

Uploads and original downloads

JSON, exit codes and configuration

CLI execution path and APIs

Commands and backend operations

Requests, recovery, and automation