theta-edgecloud-skill（西塔边缘云技能）

Theta EdgeCloud Skill (Cloud API Runtime)

Dedicated inference status (revalidated 2026-03-11)

• Dedicated OpenAI-compatible inference (theta.inference.models, theta.inference.chat) was revalidated successfully after Developer Plan / quota upgrade.
• Important readiness nuance: endpoint creation is not instantly probe-ready.
• Observed post-upgrade warm-up pattern during successful live validation:

- early authenticated GET /v1/models returned transient 404, then transient 502, then succeeded - authenticated POST /v1/chat/completions succeeded after warm-up

• Operational guidance:

- use authenticated readiness retries for ~1-2 minutes before declaring dedicated endpoint failure - prefer vm_gt1 first when allocator capacity is available - if vm_gt1 is capacity-blocked, fall back to vm_gt2 or V100-backed options

Credential scope model (important)

This skill is command-scoped: only provide the credentials needed for the command family you use.

• Controller/deployment/project commands: THETA_EC_API_KEY, THETA_EC_PROJECT_ID
• Balance command: add THETA_ORG_ID
• On-demand inference commands: THETA_ONDEMAND_API_TOKEN or THETA_ONDEMAND_API_KEY
• Inference endpoint commands: THETA_INFERENCE_ENDPOINT + auth (THETA_INFERENCE_AUTH_TOKEN or user/pass)

Credentials above are not globally required all at once.

Quick setup (new users)

1) Log in at https://www.thetaedgecloud.com/. 2) Go to Account -> Projects and select your project. 3) Click Create API Key and copy the key. 4) Provide these on install/setup prompt: - THETA_EC_API_KEY - THETA_EC_PROJECT_ID 5) (Recommended for on-demand image/video generation) create On-demand API key/token and set: - THETA_ONDEMAND_API_KEY (or THETA_ONDEMAND_API_TOKEN) 6) For dedicated deployments / GPU-backed serving, check Account -> Quota and click Increase Quota if needed. - Default machine-type quotas may be too low (or zero) for dedicated GPU deployment creation. - Current live operator guidance: add at least $20 in credits first so the organization can reach Developer Plan, then retry Increase Quota for fuller functionality. - Higher quota tiers may still exist beyond Developer Plan.

If a command says a key is missing, run theta.auth.capabilities to see exactly what to configure.

This runtime artifact is scoped to cloud API operations only.

Security behavior (explicit)

• Runtime command handlers do not execute local shell commands.
• Runtime does not read local files for upload operations.
• Runtime does not call localhost/default local RPC endpoints.
• Dedicated inference endpoint override via args.endpoint is disabled in runtime commands; use THETA_INFERENCE_ENDPOINT.
• Runtime secret resolution uses OpenClaw secret provider first, then env fallback for:

- THETA_ONDEMAND_API_TOKEN - THETA_INFERENCE_AUTH_TOKEN - THETA_INFERENCE_AUTH_USER / THETA_INFERENCE_AUTH_PASS

• Paid/mutating operations are user-triggered and can be gated by THETA_DRY_RUN=1.

Credential model (what users actually need)

Users need a Theta EdgeCloud account with billing/credits enabled for paid operations.

Use only credentials required for the feature set you plan to call:

• Deployments API:

- THETA_EC_API_KEY - THETA_EC_PROJECT_ID

• Dedicated inference endpoint:

- THETA_INFERENCE_ENDPOINT - EITHER basic auth: - THETA_INFERENCE_AUTH_USER - THETA_INFERENCE_AUTH_PASS - OR bearer token auth: - THETA_INFERENCE_AUTH_TOKEN

• On-demand model API:

- THETA_ONDEMAND_API_TOKEN

• Theta Video API:

- THETA_VIDEO_SA_ID - THETA_VIDEO_SA_SECRET

Runtime-only package

This ClawHub artifact is a dist/docs bundle intended for transparent inspection and low scanner surface.

Env knobs (selected)

• THETA_DRY_RUN
• THETA_EC_API_KEY
• THETA_EC_PROJECT_ID
• THETA_ORG_ID
• THETA_INFERENCE_ENDPOINT
• THETA_INFERENCE_AUTH_USER
• THETA_INFERENCE_AUTH_PASS
• THETA_INFERENCE_AUTH_TOKEN
• THETA_ONDEMAND_API_TOKEN
• THETA_ONDEMAND_API_KEY
• THETA_API_KEY (MCP-compatible on-demand alias)
• THETA_VIDEO_SA_ID
• THETA_VIDEO_SA_SECRET
• THETA_HTTP_TIMEOUT_MS
• THETA_HTTP_MAX_RETRIES
• THETA_HTTP_RETRY_BACKOFF_MS

On-demand service status (refreshed 2026-04-25 for v0.1.21)

Live public service discovery currently exposes these aliases:

• Chat/LLM: qwen3, minimax_m2_5, gpt_oss_120b, llama_3_1_70b, llama_3_8b
• Image/video/vision/audio: flux, stable_diffusion_xl_turbo, step_video, grounding_dino, blip, llava, whisper

Qwen3 notes:

• Canonical slug: qwen3
• Request family: chat/completions, not simple prompt text
• Payload shape: input.messages = [{ role, content }]
• Runtime command: theta.ondemand.chat, default service qwen3
• Runtime default for Qwen3 chat: stream:true with an internal SSE parser, because live Qwen3 is much more likely to respond through Theta's streaming path than the non-streaming JSON path.
• Observed variant: parallax_32b_fp8
• Capacity can temporarily return 409 No instances available - try again later; v0.1.21 treats this as retriable temporary capacity exhaustion.

AI Services coverage

• Deployments API: list + create + stop + delete
• Dedicated model templates: standard + custom
• On-demand model APIs: live discovery + infer/status/poll/chat
• Dedicated inference endpoint: models + chat
• Dedicated deployments listing
• Jupyter notebook listing
• GPU node and GPU cluster listing
• Persistent storage listing
• Agentic AI (chatbot) listing
• Theta Video APIs: list/upload/video/stream/ingestor operations

Theta-only OpenClaw operating options (no other subscriptions)

If Theta is the only paid AI backend, this skill can still cover most OpenClaw execution routes:

• Content generation:

- image/logo/creative generation (flux, stable_diffusion_*) via theta.ondemand.infer - image enhancement/upscale (esrgan) - identity-preserving generation (instant_id) - virtual try-on/product visualization (stable_viton) - video generation (step_video) and talking avatars (talking_head)

• Website AI features:

- chatbot/support/Q&A/rewrite pipelines using on-demand LLMs (qwen3, minimax_m2_5, gpt_oss_120b, llama_3_8b, llama_3_1_70b)

• Vision/media intelligence:

- captioning/alt-text (blip), object detection (grounding_dino), vision-language Q&A (llava), transcription (whisper)

• Video infrastructure:

- upload/video/stream/ingestor operations via theta.video.*

• Compute/ops:

- VM/deployment lifecycle + GPU/storage listings + capability/balance checks via theta.deployments.*, theta.ai.*, theta.auth.capabilities, theta.billing.balance

Recommended reliability route:

• Prefer on-demand + video/controller flows for production automation.
• Current validated daily-use routes include flux image generation, step_video on-demand video generation, live on-demand service discovery, chat/completions through the Qwen3-compatible adapter, and dedicated OpenAI-compatible inference after readiness warm-up.
• Dedicated endpoint commands are now valid when the project has quota/plan support, but should use authenticated readiness retries instead of immediate fail-fast assumptions.

Organization & Project scope

• Theta dashboard uses Organization + Project context.
• Runtime commands are project-scoped and require explicit projectId where relevant.
• Org membership/invite/session management endpoints are web-dashboard auth flows and not included in this skill runtime.

API key vs user/password auth (validated)

• API key (THETA_EC_API_KEY) + project/org IDs can access project-scoped controller APIs and org balance.
• API key is sufficient for key runtime operations; dashboard username/password is not required for these flows.
• Username/password session auth is still required for account-management endpoints (org/project membership, invite, charge usage history APIs).

On-demand API key alias

The runtime accepts either THETA_ONDEMAND_API_TOKEN or THETA_ONDEMAND_API_KEY for on-demand model API auth.

Reliability behavior

• theta.ai.dedicatedDeployments.list now emits warning metadata if serving-template catalog calls fail; this avoids silent empty output.
• theta.ondemand.listServices returns service entries tagged by source (live or catalog) and includes fallback reason/warning fields when live discovery is unavailable.
• step_video now uses service-aware timeout behavior:

- submit timeout is automatically raised above generic HTTP defaults - poll/completion timeout auto-scales from requested video size when the request status exposes frame/fps metadata - sizing rounds up with extra variance buffer to reduce false local timeouts under load

Auth diagnostics

Use theta.auth.capabilities to quickly see which command families are available with the current credential set and which env vars are missing.

First-run setup command

Use theta.setup to get a one-screen checklist for new users, including where to create API keys and which env vars to set.