eachlabs

name	eachlabs
description	Use when building, debugging, or explaining integrations with each::labs, eachlabs.ai, each::api, each::workflows, each::sense, the each::labs LLM Router, or the each::labs MCP. Also use for image, video, audio, 3D, or workflow generation tasks that mention Eachlabs or EACHLABS_API_KEY.

each::labs

Use this skill to build practical each::labs integrations in Codex. Prefer current official documentation over memory when endpoint details, model slugs, schemas, or pricing behavior matter.

Core Surfaces

• each::api: direct REST model execution at https://api.eachlabs.ai.
• each::workflows: multi-step pipelines at https://workflows.eachlabs.run/api/v1.
• each::sense: OpenAI-compatible generative media agent at https://eachsense-agent.core.eachlabs.run.
• LLM Router: OpenAI-compatible LLM access at https://api.eachlabs.ai/v1.
• MCP: bundled eachlabs server with docs proxy tools plus action tools.

Docs Workflow

When the bundled eachlabs MCP is available, use it for current docs lookup before implementing or answering precise API questions:

1. Search with search_each_labs.
2. Read exact pages with query_docs_filesystem_each_labs.
3. Cite or link the public docs URL when reporting user-facing API facts.

If MCP tools are unavailable, use the public docs index at https://docs.eachlabs.ai/llms.txt.

Prefer its purpose-built tools over hand-written HTTP calls:

• eachlabs_search_models, eachlabs_get_model, and eachlabs_get_model_request_schema for model discovery and schemas.
• eachlabs_recommend_models followed by eachlabs_compare_models for credit-free model selection using live schemas and catalog p50 latency.
• eachlabs_diff_model_schema before upgrades and deploys to detect breaking live-schema drift.
• eachlabs_create_prediction (mode async, wait, or sync; validates input locally first) and eachlabs_get_prediction (wait=true to poll) for direct generation; successful image outputs come back inline in chat.
• eachlabs_list_executions for raw run history, eachlabs_summarize_usage for privacy-conscious cost/runtime summaries, eachlabs_diagnose_run for local failure triage, and eachlabs_export_debug_bundle for shareable diagnostics without sensitive payloads.
• eachlabs_upload_file for local media inputs.
• eachlabs_audio_transcribe and eachlabs_audio_speech for the dedicated audio APIs.
• eachlabs_create_workflow, eachlabs_execute_workflow, and workflow execution polling tools for workflows.
• eachlabs_validate_workflow_definition before workflow mutation, eachlabs_diff_workflow before publishing a version, and eachlabs_generate_integration_code for live-schema TypeScript, Python, Go, or cURL examples.
• eachsense_chat_completion and eachsense_build_workflow for each::sense.
• eachlabs_llm_chat_completion for LLM Router calls.

Credential Rules

• Never print, log, paste, or summarize EACHLABS_API_KEY.
• Never print, log, paste, or summarize EACH_API_KEY.
• Use EACH_API_KEY or EACHLABS_API_KEY for local server-side calls.
• Do not put Eachlabs keys in client-side code, public repositories, screenshots, or build output.
• For current direct REST APIs, authenticate with Authorization: Bearer.
• Keep X-API-Key only for explicitly documented legacy compatibility.
• Public model listing, details, schemas, recommendations, comparisons, validation, and code generation work without credentials. Live generation and account-specific endpoints need a key.

Choosing The Right Surface

• Use each::sense when the user wants natural-language media generation, model auto-selection, agent-style orchestration, or multi-turn media work.
• Use each::api when the user names a specific model, needs deterministic schema control, needs webhooks, or wants direct polling of prediction status.
• Use each::workflows when the request involves repeated pipelines, multiple model steps, fallback behavior, branching, bulk runs, or reusable workflow versions.
• Use the LLM Router for text-only LLM routing with OpenAI-compatible clients.
• Use the docs MCP for documentation discovery, not for generation or account operations.
• For model choice, compare live candidates before spending credits; catalog p50 is a latency signal, not a price or quality guarantee.

Direct Prediction Pattern

For each::api direct model execution:

1. Fetch or confirm the model schema before constructing inputs.
2. Send POST /v1/prediction with model, version, and input.
3. Poll GET /v1/prediction/{id} until success, failed, error, or cancelled.
4. Prefer webhooks for production integrations that can receive callbacks.
5. Surface model cost, status, output URL, and logs when available.

For a failed run, call eachlabs_diagnose_run before retrying. A dispatched paid write can have an ambiguous result; check execution history before creating replacement work.

The default prediction status flow is starting to processing to a terminal state.

Safety And Media Handling

• The safety checker is enabled by default.
• Only pass enable_safety_checker: false when the user explicitly asks for that behavior and the selected model supports it.
• For each::sense, pass safety-checker options as top-level request fields rather than inside a model input object.
• Treat generated media URLs as potentially user-visible artifacts; avoid leaking private prompts, credentials, or webhook secrets.
• Inline media fetches accept HTTPS only and reject localhost, private IPs, internal hostnames, and unsafe redirects. Keep extensionless signed URLs as resource links when their MIME type cannot be safely confirmed.
• The raw request escape hatch must not perform unauthenticated writes.

Implementation Standards

• Keep examples server-side unless the user explicitly asks for a browser-only prototype.
• Validate generated request bodies against the model request_schema when practical.
• Add focused tests around payload construction, status polling, error handling, and webhook signature verification.
• For local smoke tests without credentials, prefer GET https://api.eachlabs.ai/v1/models?limit=3.
• For live generation tests, ask before spending credits or creating external artifacts.

Common Environment

EACHLABS_API_KEY=...
EACHLABS_API_BASE_URL=https://api.eachlabs.ai
EACHSENSE_BASE_URL=https://eachsense-agent.core.eachlabs.run
EACHLABS_WORKFLOWS_BASE_URL=https://workflows.eachlabs.run/api/v1