sac

Use sac (SeaArt CLI) to generate images, video, audio, and 3D assets, chat with LLMs, inspect task results, or manage SeaArt API credentials and config.

SeaVerseAI 1 Updated 1mo ago

Resources

GitHub

Install

npx skillscat add seaverseai/sea-cli

Install via the SkillsCat registry.

SKILL.md

sac Skill

Operational reference for the binary-only public mirror. Read this after CLAUDE.md or AGENTS.md.

Install

Run from the repository root (the directory containing install.sh):

bash install.sh
source ./.sac-env
sac --version

The installer writes two files in the repository root:

.sac-bin-path — absolute path to the installed binary (e.g. /root/.local/bin/sac)
.sac-env — a single export PATH=... line for the install directory

source ./.sac-env only affects the current shell process. If sac is still not found after sourcing, read the absolute path and invoke it directly:

SAC_BIN="$(cat .sac-bin-path)"
"$SAC_BIN" --version

In all subsequent shell commands in this session, if PATH propagation is uncertain, substitute "$(cat .sac-bin-path)" for bare sac.

To install to a custom directory:

SAC_INSTALL_DIR=/usr/local/bin bash install.sh

Authentication

Required before any generate or chat command.

Save a key to ~/.sac/config.json (validates before saving):

sac auth login --api-key sa-xxxxxxxx

Or pass via environment variable (takes precedence over config file, no file write):

export SAC_API_KEY=sa-xxxxxxxx

Verify:

sac auth status          # show stored key info (no network call)
sac auth status --check  # validate key against the API (network call)

Validation behavior:

sac auth login validates the key before saving — a stored key is always valid at save time
401/403 from --check means the key is invalid
Task not found response means the key was accepted (probe task simply does not exist)
network error, timeout, or 5xx makes --check exit with code 6 (uncertainty, not confirmed invalid)

auth status --check exit codes:

0 — key looks valid
3 — authentication failed (key rejected)
6 — network or verification uncertainty

Remove stored key:

sac auth logout

auth login, auth status, and auth logout all support --output json.

auth status --output json (authenticated):

{
  "authenticated": true,
  "key": "sa-a...xxxx",
  "source": "config file",
  "config": "/Users/you/.sac/config.json"
}

auth status --output json (not authenticated):

{
  "authenticated": false,
  "message": "Not authenticated.",
  "hint": "Run: sac auth login --api-key <token>\nOr set: export SAC_API_KEY=<token>"
}

auth login --output json (success):

{
  "authenticated": true,
  "saved": true,
  "key": "sa-a...xxxx",
  "config": "/Users/you/.sac/config.json",
  "verification": {
    "status": "valid",
    "message": "API key accepted; probe task was not found, as expected.",
    "http_status": 404
  }
}

Agent Baseline

Use this baseline for all structured commands:

sac --non-interactive --quiet --output json <command> [flags]

--non-interactive — fail immediately instead of prompting for missing args
--quiet — suppress spinners, progress lines, and info messages on stderr
--output json — machine-readable stdout

Text-only modes reject --output json with a usage error — do not pass it to:

chat --stream
interactive chat (no --message)
generate task --output-only-url
update

Model Discovery Workflow

When the user asks about available models, or wants to call a specific model, always follow this workflow — do not guess model names or flags:

sac model search — discover models by type, provider, or modality
sac model get <model-id> — inspect the Body Template and full field reference (capture silently into a variable)
sac generate submit --body-json '...' — call the model with exact parameters from the template

# Example: user wants an image-to-video model from Kling
sac model search --type i2v --provider kling --output json
# → pick model id from results, e.g. kling_v3_i2v

MODEL_SKILL=$(sac model get kling_v3_i2v)
# → $MODEL_SKILL contains Body Template and Fields; never print it

sac generate submit --body-json '{
  "model": "kling_v3_i2v",
  "dash_scope": true,
  "moderation": true,
  "input": [{"params": {"prompt": "...", "image": "https://..."}}],
  "metadata": {}
}'

Use generate image / video / audio / 3d directly only when the model and flags are already known with certainty. When in doubt, use the discovery workflow above.

Flag	Description
`--api-key <token>`	Temporary bearer token for this command only; does not persist. Exception: `auth login --api-key` validates and saves it
`--base-url <url>`	Temporary gateway base URL for this command only; does not persist
`--output json\|text`	Force output format (default: auto-detect by TTY)
`--timeout <seconds>`	Request timeout (default: 300)
`--quiet`	Suppress spinners and info messages
`--verbose`	Print HTTP request/response details to stderr
`--dry-run`	Show what would be sent without executing
`--non-interactive`	Fail on missing args instead of prompting
`--async`	Return task ID immediately without polling
`--no-color`	Disable ANSI colors

One-off temporary overrides:

sac --api-key sa-xxxxxxxx generate image --prompt "..."
sac --base-url https://gateway.example.com generate image --prompt "..."

auth login --api-key is different: it validates the token and writes it to persisted config. The temporary meaning only applies to non-login commands.

--base-url accepts a gateway root (https://gateway.example.com) or explicit /model / /llm base URLs. It only affects the current command and never writes config.

Configuration

Priority order (highest to lowest):

command-line --flag
environment variable
~/.sac/config.json
built-in default

Key	Env var	Default
`api_key`	`SAC_API_KEY`	—
`output`	`SAC_OUTPUT`	auto-detect
`timeout`	`SAC_TIMEOUT`	`300`
`default_image_model`	—	`volces_seedream_4_5`
`default_chat_model`	—	`deepseek-v3-0324`
`base_url`	`SAC_BASE_URL`	SeaArt gateway

sac config show                                      # view current config
sac config show --output json
sac config set --key default_chat_model --value deepseek-v3-0324
sac config set --key output --value json
sac config set --key timeout --value 600

config show --output json:

{
  "base_url": "https://openresty-gateway.api.seaart.ai",
  "output": "json",
  "timeout": 300,
  "config_file": "/Users/you/.sac/config.json",
  "api_key": "sa-a...xxxx",
  "default_chat_model": "deepseek-v3-0324"
}

Valid config set keys: api_key, output, timeout, default_image_model, default_chat_model, base_url

Output Semantics

TTY → human-readable text with spinners
Non-TTY / pipe → JSON automatically when the command supports structured output
--output json or SAC_OUTPUT=json forces JSON in any context

Command Reference

generate image

Create images. Polls until complete and returns URLs by default.

sac generate image --prompt "a cat in space"
sac generate image --prompt "product shot" --model kling_v3_image --aspect-ratio 1:1
sac generate image --prompt "国风山水画" --model tencent_image_creation_3 --resolution 1024:1024 --logo-add 1 --revise 1
sac generate image --prompt "make it anime" --image-url https://example.com/input.png --action 1
sac generate image --prompt "text" --n 2
sac generate image --prompt "text" --width 1024 --height 1024
sac generate image --prompt "text" --out-dir ./output
sac generate image --prompt "text" --async
sac generate image --list-models
sac generate image --list-models --provider volces

Flags:

Flag	Type	Default	Description
`--prompt <text>`	string	required	Image description
`--model <id>`	string	`volces_seedream_4_5`	Model ID (overridden by `default_image_model` config if set)
`--n <count>`	number	1	Number of images
`--width <px>`	number	—	Image width
`--height <px>`	number	—	Image height
`--aspect-ratio <ratio>`	string	—	e.g. `16:9`, `1:1`
`--resolution <preset>`	string	—	e.g. `1024:1024`
`--seed <n>`	number	—	Random seed
`--steps <n>`	number	20	Inference steps
`--cfg-scale <n>`	number	—	Guidance scale
`--negative-prompt <text>`	string	—	Negative prompt
`--image-url <url>`	string	—	Input image for i2i
`--action <n>`	number	0	0=t2i, 1=i2i, 3=t2i+controlnet, 5=i2i+controlnet
`--out-dir <dir>`	string	—	Download images to this directory
`--out-prefix <prefix>`	string	`image`	Filename prefix for downloads
`--async`	boolean	false	Return task ID without polling
`--list-models`	boolean	—	List available models and exit
`--provider <id>`	string	—	Filter `--list-models` by provider

Built-in default model: volces_seedream_4_5

SeaArt models: sdxl, z_image, z_image_turbo

Volces models: volces_seedream_5, volces_seedream_4_5, volces_seedream_4_0, volces_seedream_3_0, volces_seedream_4_5_multi_blend, volces_jimeng_3_1, volces_jimeng_3_0, volces_jimeng_i2i_3_0, volces_jimeng_tilesr, volces_seededit_3_0, volces_seededit_3_0_i2i, volces_seededit_single_ip, volces_seededit_multi_ip, volces_seededit_multi_style, volces_seededit_3d_style, volces_seededit_portrait

Alibaba models: alibaba_wan27_image_pro

Nano models: nano_banana_2

Kling models: kling_v3_image, kling_omni_image, kling_v3_omni_image

Tencent models: tencent_image_creation_3 (supports --resolution, --seed, --logo-add 0|1, --revise 0|1)

--output json (polling complete, no --out-dir):

{ "task_id": "d7erdule878c738sr94g", "urls": ["https://image.cdn2.seaart.me/..."] }

--output json with --out-dir:

{ "task_id": "d7erdule878c738sr94g", "saved": ["./output/image_001.webp"] }

--output json with --async:

{ "task_id": "d7erdule878c738sr94g", "status": "in_progress" }

generate video

Create videos. Polls until complete and returns URLs by default.

sac generate video --prompt "a fox in snow"
sac generate video --prompt "text" --model vidu_q3_pro --async
sac generate video --prompt "text" --model kling_v3 --aspect-ratio 16:9 --duration 5
sac generate video --prompt "car drives away" --model kling_v3_i2v --image-url https://example.com/img.webp
sac generate video --model kling_avatar --image-url https://example.com/avatar.png --audio-url https://example.com/voice.mp3
sac generate video --prompt "你好" --model kling_lipsync --lipsync-mode text2video --video-url https://example.com/face.mp4 --voice-id voice_123 --voice-language zh
sac generate video --model tencent_mps_super_resolution --video-url https://example.com/input.mp4 --resolution 1080P
sac generate video --prompt "commercial" --model vidu_q3_reference --image-urls https://example.com/img.webp
sac generate video --prompt "commercial" --model alibaba_wanx26_reference --reference-urls https://example.com/img.jpg
sac generate video --prompt "a fox in snow" --model volces_seedance_3_0 --resolution 720p
sac generate video --model volces_seedance_30_i2v --image-url https://example.com/first.jpg --prompt "slow push"
sac generate video --prompt "cinematic city at night" --model minimax_hailuo_02 --duration 10 --resolution 768P --prompt-optimizer false
sac generate video --model minimax_hailuo_23_fast_i2v --image-url https://example.com/first.jpg --prompt "gentle push in"
sac generate video --list-models

Built-in default model: volces_seedance_1_5_pro

Vidu models: vidu_q3_pro (t2v), vidu_q3_pro_i2v (i2v, requires --image-url), vidu_q3_reference (requires --image-urls), vidu_q3_pro_start_end (requires --image-url start + --image-tail-url end), vidu_template_v2 (requires --image-url, --template), vidu_one_click_mv (requires --image-urls, --audio-url)

Kling models:

T2V: kling_v1, kling_v1_5, kling_v1_6, kling_v2_master, kling_v2_1_master, kling_v2_5_turbo, kling_v2_6, kling_v3
I2V: kling_v1_i2v, kling_v1_5_i2v, kling_v1_6_i2v, kling_v2_1_i2v, kling_v2_master_i2v, kling_v2_1_master_i2v, kling_v2_5_turbo_i2v, kling_v2_6_i2v, kling_v3_i2v
Avatar: kling_avatar (requires --image-url + --audio-url or --audio-id)
Motion control: kling_motion_control, kling_v3_motion_control (require --image-url, --video-url, --character-orientation image|video, --mode std|pro)
Effects single: kling_effects_single (requires --image-url, --effect-scene, --duration)
Effects multi: kling_effects_multi_v1, kling_effects_multi_v15, kling_effects_multi_v16 (require exactly 2 --image-urls, --effect-scene, --duration)
Duration extension: kling_duration_extension (requires --video-url, --duration)
Lipsync: kling_lipsync — --lipsync-mode text2video (requires --prompt, --video-url/--video-id, --voice-id, --voice-language) or --lipsync-mode audio2video (requires --audio-url)
Omni: kling_omni_video, kling_v3_omni_video (supports --image-url, --video-url, --aspect-ratio, --duration)

Tencent models: tencent_kling_v3, tencent_kling_v3_omni, tencent_mps_super_resolution (requires --video-url, --resolution 720P|1080P|2K|4K, optional --short 0|1)

Alibaba models: alibaba_wanx26_t2v (t2v), alibaba_wanx26_i2v (i2v, requires --image-url, supports --resolution 720P|1080P), alibaba_wanx26_reference (requires --reference-urls; URLs must be accessible from China — avoid Google Cloud Storage CDN)

Volces models: volces_seedance_1_5_pro, volces_seedance_2_0, volces_seedance_2_0_fast, volces_seedance_3_0, volces_seedance_3_0_pro, volces_seedance_30_i2v, volces_seedance_pro_fast, volces_draft_video, volces_jimeng_dream_actor_m1, volces_jimeng_dream_actor_m2, volces_realman_avatar_picture_omni_v2, volces_realman_avatar_picture_omni_v15, volces_realman_avatar_imitator_v2v

Key flags: --duration, --size, --aspect-ratio, --resolution, --seed, --fps, --frames, --image-url, --image-tail-url, --image-urls, --video-url, --video-id, --audio-url, --audio-id, --voice-id, --voice-language, --voice-speed, --lipsync-mode, --effect-scene, --character-orientation, --keep-original-sound, --extension-type, --template, --reference-urls, --mask-urls, --audio (AI audio), --shot-type (Wanx t2v)

--output json (polling complete):

{ "task_id": "...", "urls": ["https://..."] }

generate audio

sac generate audio --prompt "epic orchestral theme"
sac generate audio --prompt "pop anthem" --model mureka_song_generator --lyrics "verse one..."
sac generate audio --model kling_video_to_audio --video-url https://example.com/clip.mp4 --sound-effect-prompt "rain and city"
sac generate audio --prompt "indie folk, wistful, warm guitar" --model minimax_music_25_plus --instrumental --format wav
sac generate audio --prompt "你好，这是一段旁白" --model minimax_t2a --voice-id female-chengshu --voice-speed 1.1 --output-format url

Built-in default model: lyria_3_pro_preview

Models:

lyria_3_pro_preview — music from text prompt; requires --prompt
mureka_song_generator — song with lyrics; requires --lyrics; optional --prompt, --mureka-model, --n, --reference-id, --vocal-id, --melody-id
kling_video_to_audio — generate audio for a video; requires --video-url or --video-id; optional --sound-effect-prompt, --n
minimax_music_25, minimax_music_25_plus — MiniMax music generation; supports --lyrics, --lyrics-optimizer, --instrumental, --sample-rate, --bitrate, --format
minimax_music_generation — MiniMax music generation with required --minimax-model
minimax_t2a — MiniMax text-to-speech; supports --minimax-model, --voice-id, --voice-speed, --voice-volume, --voice-pitch, --voice-emotion, --sample-rate, --bitrate, --format, --channel, --output-format

--output json (polling complete):

{ "task_id": "...", "urls": ["https://..."] }

generate 3d

sac generate 3d --prompt "a stylized toy robot"
sac generate 3d --image-url https://example.com/object.png
sac generate 3d --image-urls https://example.com/front.png --image-urls https://example.com/left.png --image-urls https://example.com/back.png --image-urls https://example.com/right.png
sac generate 3d --model tencent_hunyuan_3d --prompt "a carved jade dragon" --result-format GLB --enable-pbr
sac generate 3d --model tencent_hunyuan_3d_pro --image-url https://example.com/obj.png --face-count 80000 --generate-type LowPoly --polygon-type triangle
sac generate 3d --model volces_seed3d --prompt "a ceramic cat" --image-url https://example.com/cat.png
sac generate 3d --list-models

Default model routing (Tripo3D):

prompt only → tripo3d_text_to_model
single --image-url → tripo3d_image_to_model
repeated --image-urls → tripo3d_multiview_to_model
mixed or Tencent/Volces inputs → pass --model explicitly

Volces: volces_seed3d (requires --prompt + --image-url)

Tencent:

tencent_hunyuan_3d — requires exactly one of --prompt/--image-url/--image-base64; supports --result-format, --enable-pbr, repeated --multi-view-image key=url
tencent_hunyuan_3d_pro — same input; supports --face-count, --generate-type, --polygon-type, --enable-pbr
tencent_hunyuan_3d_rapid — same input; supports --result-format, --enable-pbr

--output json (polling complete):

{ "task_id": "...", "urls": ["https://..."] }

generate task

Query or wait for any generation task by ID.

sac generate task <task-id>
sac generate task <task-id> --output json
sac generate task <task-id> --output-only-url   # raw URL lines, rejects --output json

--output json:

{
  "id": "d7erdule878c738sr94g",
  "status": "completed",
  "progress": 1,
  "output": [
    {
      "content": [
        { "type": "image", "url": "https://image.cdn2.seaart.me/..." }
      ]
    }
  ]
}

Download URLs are at output[].content[].url.

chat

Automation rule: always pass --message and --non-interactive from scripts or agents. Without --message in a TTY, an interactive REPL starts and blocks on keyboard input. Without --message outside a TTY, the command exits with code 2 instead of hanging.

sac chat --message "Hello" --non-interactive --quiet --output json
sac chat --message "Hello" --stream                  # streaming tokens (text only)
sac chat --model gemini-2.5-pro --message "Hello"
sac chat --system "You are a code reviewer." --message "Review this: ..."
sac chat --message "user:Hi" --message "assistant:Hello" --message "Why?"
sac chat --messages-file messages.json
sac chat --messages-file -                           # read messages array from stdin
sac chat models
sac chat models --filter claude --output json
sac chat set-model --model deepseek-v3-0324

Default model: deepseek-v3-0324. Override with --model or persist with sac chat set-model.

Flags:

Flag	Type	Default	Description
`--model <id>`	string	`deepseek-v3-0324`	Model ID
`--message <text>`	array	—	Message text (repeatable); prefix `role:` to set role
`--messages-file <path>`	string	—	JSON messages array file; `-` reads stdin
`--system <text>`	string	—	System prompt
`--max-tokens <n>`	number	—	Max tokens to generate
`--temperature <n>`	number	—	Sampling temperature
`--stream`	boolean	auto	Force streaming output

Message role syntax:

--message "user:What is 2+2?"
--message "assistant:4"
--message "Why?"              # defaults to user role

Messages file format:

[
  { "role": "system", "content": "You are a helpful assistant." },
  { "role": "user", "content": "Hello" }
]

--output json (non-streaming single-turn):

{
  "id": "abc123",
  "model": "deepseek-v3-0324",
  "choices": [
    {
      "message": { "role": "assistant", "content": "Hello! How can I help?" },
      "finish_reason": "stop"
    }
  ],
  "usage": { "prompt_tokens": 4, "completion_tokens": 9, "total_tokens": 13 }
}

chat models --output json:

{ "models": ["deepseek-v3-0324", "deepseek-r1", "claude-sonnet-4-6", "..."] }

generate submit

Submit a raw request body directly to the generation API. Use this when generate image/video/audio/3d flags don't cover the params you need, or after model get to call a model with full parameter control.

sac generate submit --body-json '{"model":"kling_v3_i2v","dash_scope":true,"moderation":true,"input":[{"params":{"prompt":"...","image":"https://..."}}],"metadata":{}}'
sac generate submit --body-json '{"model":"comfyuifast","input":[{"params":{"prompt":"1 girl"}}]}' --async
sac generate submit --body-json '...' --out-dir ./output

Flags:

Flag	Type	Description
`--body-json <json>`	string	Complete request body as a JSON string (required)
`--out-dir <dir>`	string	Download output files to this directory
`--out-prefix <prefix>`	string	Filename prefix for downloads (default: `output`)
`--async`	boolean	Return task ID immediately without polling

model search

Search models (backed by Meilisearch).

sac model search                                           # list all
sac model search --query kling                             # full-text search
sac model search --input-modality image --output-modality video   # filter by i/o modality
sac model search --type i2v                                # filter by tag abbreviation
sac model search --type i2v --provider alibaba             # combined filter
sac model search --query wan --input-modality image --output-modality video
sac model search --output-modality audio                   # all audio output models
sac model search --output json                             # machine-readable

Filter flags (all repeatable for AND semantics):

Flag	Values	Description
`--query <text>`	any	Full-text search across name, description, skill content
`--input-modality <m>`	`text` `image` `video` `audio`	Input modality filter
`--output-modality <m>`	`image` `video` `audio` `3d`	Output modality filter
`--type <abbr>`	see table below	Tag abbreviation filter
`--provider <name>`	`kling` `alibaba` `volces` `vidu` `pixverse` `tencent` `tripo3d` `google` `mureka` `seaart`	Provider filter
`--limit <n>`	number	Max results (default: 20)

Tag abbreviations (--type):

Abbr	Full name	中文
`t2i`	text-to-image	文生图
`i2i`	image-to-image	图生图
`t2v`	text-to-video	文生视频
`i2v`	image-to-video	图生视频
`r2v`	reference-to-video	参考生视频
`t2a`	text-to-audio	文生音频
`a2a`	audio-to-audio	音频生音频
`t23d`	text-to-3d	文生3D
`i23d`	image-to-3d	图生3D
`fx`	effects-template	特效模板
`vext`	video-extension	视频延长
`isr`	image-super-resolution	图像超分
`vsr`	video-super-resolution	视频超分
`vref`	video-reference	视频参考
`lipsync`	lipsync	对口型
`v2a`	video-to-audio	视频配乐
`vedit`	video-editing	视频编辑
`lyrics`	lyrics-generation	歌词生成

--output json response:

{
  "estimatedTotalHits": 12,
  "hits": [
    {
      "id": "kling_v3_i2v",
      "name": "kling_v3_i2v",
      "provider": "kling",
      "description": "Kling AI V3 图生成视频接口",
      "input": ["audio", "image", "text"],
      "output": ["video"],
      "media_type": "video",
      "tags": ["image-to-video"],
      "tags_abbr": ["i2v"]
    }
  ]
}

model get

Get the full SKILL.md for a model (body template + fields reference).

Silent-capture rule: always assign output to a variable — never print model get output to the terminal.

MODEL_SKILL=$(sac model get kling_v3_i2v)
MODEL_SKILL=$(sac model get alibaba_wanx26_t2v)

Returns raw SKILL.md content to stdout. Use with generate submit for full parameter control:

# Discover → inspect → call workflow
sac model search --type i2v --provider kling --output json \
  | node -e "const d=JSON.parse(require('fs').readFileSync(0,'utf8')); console.log(d.hits[0].name)"
# → kling_v3_i2v

MODEL_SKILL=$(sac model get kling_v3_i2v)
# → $MODEL_SKILL contains Body Template and required/optional Fields; never print it

sac generate submit --body-json '{
  "model": "kling_v3_i2v",
  "dash_scope": true,
  "moderation": true,
  "input": [{"params": {"prompt": "...", "image": "https://..."}}],
  "metadata": {}
}'

Async Task Flow

For long-running generation tasks, use --async to get the task ID immediately, then poll separately:

# Step 1: submit and capture task ID
RESULT=$(sac generate image --prompt "a cat" --async --quiet --output json)
TASK_ID=$(echo "$RESULT" | node -e "process.stdin.setEncoding('utf8'); let d=''; process.stdin.on('data',c=>d+=c); process.stdin.on('end',()=>console.log(JSON.parse(d).task_id))")

# Step 2: poll until complete
sac generate task "$TASK_ID" --output json

Or use --out-dir to download assets directly after polling completes:

sac generate image --prompt "a cat" --out-dir ./output --output json
# returns: { "task_id": "...", "saved": ["./output/image_001.webp"] }

Piping Patterns

Extract the first URL from image output:

sac generate image --prompt "a cat" --quiet --output json \
  | node -e "const d=JSON.parse(require('fs').readFileSync(0,'utf8')); console.log(d.urls[0])"

Send chat messages via stdin:

echo '[{"role":"user","content":"hello"}]' | sac chat --messages-file - --quiet --output json

Get first model matching a filter:

sac chat models --filter deepseek --output json --quiet \
  | node -e "const d=JSON.parse(require('fs').readFileSync(0,'utf8')); console.log(d.models[0])"

Exit Codes

Code	Meaning
0	Success
1	General error
2	Usage / missing required flag
3	Authentication failed
4	Quota exceeded
5	Timeout
6	Network or validation uncertainty
10	Content filtered

Network Proxy

Node.js fetch (Node 18+) requires uppercase proxy variables:

HTTPS_PROXY=http://127.0.0.1:9990 HTTP_PROXY=http://127.0.0.1:9990 sac generate image --prompt "..."

Lowercase https_proxy is ignored by Node.js fetch.

Configuration Precedence

--api-key flag  >  SAC_API_KEY env  >  ~/.sac/config.json  >  (error: not authenticated)
--base-url flag >  SAC_BASE_URL env  >  config file  >  built-in defaults
--output flag   >  SAC_OUTPUT env   >  config file         >  auto-detect by TTY
--timeout flag  >  SAC_TIMEOUT env  >  config file         >  300 seconds

sac

Resources

Install

sac Skill

Install

Authentication

Agent Baseline

Model Discovery Workflow

Configuration

Output Semantics

Command Reference

generate image

generate video

generate audio

generate 3d

generate task

chat

generate submit

model search

model get

Async Task Flow

Piping Patterns

Exit Codes

Network Proxy

Configuration Precedence

Categories

Install

Recommended Skills