venice-text-routing
Route a prompt to the right Venice text model based on privacy tier (anonymized / private / TEE / E2EE), modality (vision / audio / video input), capability (reasoning, code, function calling, web search, large context, structured output), and cost. Use when a local agent (Claude Code, Hermes, NanoClaw, Codex CLI, etc.) needs to decide whether to handle a prompt locally or escalate to Venice, and if escalating, which Venice text model to call. Sits one level above venice-models (model discovery) and before venice-chat (the call surface).
veniceai · 已有 AI 审核记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add veniceai/skills --skill venice-text-routing技能说明
name: venice-text-routing description: Route a prompt to the right Venice text model based on privacy tier (anonymized / private / TEE / E2EE), modality (vision / audio / video input), capability (reasoning, code, function calling, web search, large context, structured output), and cost. Use when a local agent (Claude Code, Hermes, NanoClaw, Codex CLI, etc.) needs to decide whether to handle a prompt locally or escalate to Venice, and if escalating, which Venice text model to call. Sits one level above venice-models (model discovery) and before venice-chat (the call surface).
Venice Text-Model Routing
This skill encodes the decision logic for "which Venice text model do I call?" — the routing layer that sits above venice-chat (the call surface) and consumes venice-models (the discovery API).
Primary use case: a local agent receives a prompt, decides whether the local model can handle it, and — if not — picks the cheapest Venice model that satisfies the privacy / modality / capability requirements.
Snapshot freshness
Before applying the matrix below, read
snapshots/text-routing.json. If the file is missing, orsnapshot_dateis older than 30 days, runpython scripts/refresh_routing.pyto regenerate it fromGET /models?type=text+GET /models/traits?type=text. Otherwise trust the cached file — do not hit/modelson every routing decision.
refresh_routing.py requires VENICE_API_KEY in the environment and rewrites both snapshots/text-routing.json and routing-matrix.md. Run it once on first install, then ~monthly (CI nightly is also fine).
When to load this skill
- Picking a Venice text model from a prompt at runtime.
- Building a local-first agent that escalates to Venice for hard prompts.
- Deciding privacy tier (anonymized vs private vs TEE vs E2EE).
- Choosing between a trait shortcut (
default_reasoning,most_intelligent, …) and a hand-filtered candidate.
For the chat call surface itself, see venice-chat. For raw discovery of every model field, see venice-models.
Privacy tier ladder
Pick the least restrictive tier that satisfies the request — restricting tier shrinks the candidate pool and often raises cost.
| Tier | Selector | Guarantee | Use when |
|---|---|---|---|
| Anonymized | model_spec.privacy: "anonymized" (no tee- / e2ee- prefix) | Venice strips IPs / metadata; the underlying provider (Anthropic, xAI, OpenAI, …) sees the prompt under their privacy policy | Default for non-sensitive workloads; only path to certain frontier closed-weights models. |
| Private | model_spec.privacy: "private" (no prefix) | Open-weights model self-hosted by Venice; zero data retention; no third party sees the prompt | Default for any user data, business logic, or anything you wouldn't paste into ChatGPT. |
| TEE | model id starts with tee- | Runs inside Intel TDX / NVIDIA confidential-compute enclave; Venice infrastructure cannot read the prompt; verifiable via GET /api/v1/tee/attestation | Regulated data, signed responses required, "Venice itself can't see this." |
| E2EE | model id starts with e2ee- | TEE + client-side ECDH (secp256k1) / HKDF-SHA256 / AES-256-GCM; prompt encrypted before leaving the device | Strongest. Healthcare, legal, secrets, anything where the wire payload must be opaque. Requires E2EE handshake — see venice-chat. |
Sources: docs.venice.ai/overview/privacy, docs.venice.ai/guides/features/tee-e2ee-models.
E2EE is not supported on /responses — route encrypted requests to /chat/completions.
Verifying a TEE claim
Two endpoints let you check that inference really ran inside an enclave. Both
are GET, unauthenticated on purpose (attestation evidence has to be
verifiable by any party without credentials), and rate limited to 10 requests
per minute per IP. Neither appears in swagger.yaml, so sync_from_swagger.py
flags /tee/attestation as stale. That is expected: the endpoints are live and
GET /models points at them from the supportsTeeAttestation capability
description.
| Endpoint | Query | Returns |
|---|---|---|
GET /api/v1/tee/attestation | model (required), nonce (optional hex, up to 64 chars / 32 bytes, used to bind an NVIDIA attestation to your challenge) | Attestation report, hardware type, and the TEE signing public key. |
GET /api/v1/tee/signature | model and request_id (required), signing_algo (optional ecdsa | ecdsa-p256 | rsa) | The provider's signature over a specific request, plus request/response hashes. |
Both return 400 if the model exists but is not TEE-attested, and 404 if the
model ID is unknown.
Verify the chain of trust in this order: fetch the attestation to get the signing public key and hardware type, confirm the recovered signer matches the attestation signing address, then verify the signature over the exact signed text the signature endpoint returned. Treat the request and response hashes as provider-reported values unless you can recompute them yourself from a documented canonical format.
Capability filters
Map prompt requirement → model_spec.capabilities flag (full list in venice-models).
| Requirement | Filter | Notes |
|---|---|---|
| Vision (single image) | supportsVision: true | Single-image vision models drop older images on each turn — chain images into the last user message. |
| Vision (multiple images) | supportsVision && supportsMultipleImages | Honor maxImages per request. |
| Audio input | supportsAudioInput: true | Audio must be base64; URLs are not accepted. |