{"schema_version":"4.2","last_updated":"2026-11-05","purpose":"Help an AI agent audit, propose, and—with explicit human approval—implement a reversible Smarty Pants trial.","service":{"name":"Smarty Pants","description":"OpenAI-compatible inference routed to approved Macs on the Smarty Pants Network.","production_origin":"https://smartypants.network","compatibility":"OpenAI chat completions"},"availability":{"status":"open_signup","public_launch_status":"early_access","message":"Anyone can sign up with an email address. New accounts get $10 of free credit, usable for 90 days. A human still approves any traffic or spending.","sign_up_url":"https://smartypants.network/sign-in","live_status_url":"https://smartypants.network/api/public/status","readiness_rule":"Check is_ready on /v1/models (or available_now from the MCP list_models tool) before sending traffic. When it is false, keep the existing provider active."},"audit_request":"Review https://smartypants.network/agents?format=json and evaluate whether Smarty Pants fits this project. Start with a read-only audit. Do not expose secrets, transmit prompt data, change production code, or create paid resources. Return our current AI cost baseline, eligible workloads with reasons, a savings range with assumptions, quality and privacy risks, and an approval-gated integration plan. If the human approves a trial, follow the implementation_playbook in that JSON; do not invent endpoints, model names, pricing, or readiness.","discovery":{"mcp_server":{"url":"https://smartypants.network/mcp","transport":"streamable-http","authentication":"none","tools":["list_models","estimate_cost","find_model","request_model","get_started"]},"signed_in_mcp_server":{"url":"https://smartypants.network/mcp/account","transport":"streamable-http","authentication":"OAuth 2.1 with PKCE; the person signs in by email and presses Allow","tools":["run_prompt","get_balance","add_credit","list_models","estimate_cost","find_model"],"spending_cap":"$20 a month per connected app; the connection lasts 90 days","payments":"add_credit returns a Stripe page that only a person pays; it charges nothing itself"},"llms_txt":"https://smartypants.network/llms.txt","agent_docs":"https://smartypants.network/docs/agents","starter_kits":["https://smartypants.network/kits/smartypants.py","https://smartypants.network/kits/smartypants.mjs"],"full_agent_flow_example":"https://smartypants.network/kits/agent-e2e.mjs","openapi":"https://smartypants.network/openapi.json","models":"https://smartypants.network/v1/models"},"model_requests":{"description":"Ask Smarty Pants to host a model that is not offered yet, or add a vote to an existing request. No sign-in needed. Requests are reviewed; nothing is hosted automatically.","list_url":"https://smartypants.network/v1/model-wishes","request":{"method":"POST","url":"https://smartypants.network/v1/model-wishes","body":{"model":"Hugging Face, OpenRouter or OpenGateway link, or a model name","note":"optional, why you need it","from":"agent"}},"vote":{"method":"POST","url":"https://smartypants.network/v1/model-wishes/{id}/vote","body":{"from":"agent"}},"page":"https://smartypants.network/request-a-model"},"instructions_for_agent":{"operating_mode":"read_only_first","safe_to_do_now":["Inspect application code and configuration names without revealing secret values.","Inventory AI providers, models, request volume, token usage, latency needs, and current costs.","Identify candidate workloads and design a quality evaluation using non-sensitive test data.","Calculate a low, expected, and high savings range with every assumption shown.","Draft an isolated provider adapter and feature-flagged rollout plan without activating it."],"human_approval_required_before":["Sending any prompt, response, file, credential, or customer data to Smarty Pants.","Changing production routing, billing, infrastructure, or deployed code.","Creating a paid resource, obtaining or storing an API key, or enabling live traffic."],"audit_steps":[{"step":"baseline","action":"Measure current usage and cost by provider and model.","return":"Monthly requests, input tokens, output tokens, spend, and p50/p95 latency."},{"step":"fit","action":"Classify workloads that smartypants/qwen3.5-4b may handle without material quality loss.","return":"Candidate workload list, exclusion reasons, sensitivity, and required evaluation cases."},{"step":"economics","action":"Apply published Smarty Pants token prices to eligible usage.","return":"Low, expected, and high monthly savings with formulas and assumptions."},{"step":"rollout","action":"Prepare a reversible integration plan.","return":"Provider adapter, feature flag, fallback route, evaluation gate, monitoring, and rollback steps."}]},"parameters":{"rate_card":{"currency":"USD","input_price_per_million_tokens":0.1,"output_price_per_million_tokens":0.25,"checked_at":"2026-10-23","note":"Confirm the live catalog before making a purchasing decision."},"limits":{"model":"smartypants/qwen3.5-4b","context_tokens":8192,"maximum_output_tokens":2048,"automatic_exclusions":["Any request whose input exceeds 8192 tokens.","Any request whose required output exceeds 2048 tokens."]},"task_taxonomy":{"likely_candidates":["Classification, tagging, extraction, and short transformations with objective expected outputs.","Templated drafting, summarization, and low-risk internal question answering that fit the published limits.","Lightweight coding or structured-output tasks with an automated correctness check."],"keep_on_current_or_premium_provider":["Requests requiring a frontier, vision, audio, embedding, or fine-tuned model.","High-stakes medical, legal, financial, safety, or security decisions.","Sensitive or regulated data unless an approved private Group security mode explicitly covers it.","Any workload that fails the team's quality or latency gate."]},"savings_formula":{"smarty_pants_cost_usd":"(eligible_input_tokens / 1000000 * 0.1) + (eligible_output_tokens / 1000000 * 0.25)","monthly_savings_low_usd":"eligible_current_cost_low_usd - smarty_pants_cost_high_usd - migration_and_fallback_cost_usd","monthly_savings_high_usd":"eligible_current_cost_high_usd - smarty_pants_cost_low_usd - migration_and_fallback_cost_usd","reporting_rule":"Return a low-to-high range with assumptions. Never turn call-volume share into spend share or return a single headline savings figure."},"example_scenario":{"label":"Example inputs only — calculate from measured team data","computers_available":50,"eligible_request_share_percent_range":[60,75],"spend_affected_share_percent_range":[10,25],"estimated_monthly_savings_usd_range":["X","Y"],"interpretation":"Eligible requests describe call volume. Spend affected is a separate, usually smaller share because suitable small-model calls are often inexpensive. The 50 Macs affect capacity and resilience; they do not prove workload fit or savings."}},"integration":{"protocol":"OpenAI-compatible chat completions","base_url":"https://smartypants.network/v1","endpoint":"/chat/completions","models_endpoint":"/models","routing_endpoint":"/smartypants/routing","model":"smartypants/qwen3.5-4b","api_key_environment_variable":"SMARTYPANTS_API_KEY","availability":"Open sign-up. An agent may prepare code now; a human creates the API key and approves activation.","example_typescript":"import OpenAI from \"openai\";\n\nconst client = new OpenAI({\n  apiKey: process.env.SMARTYPANTS_API_KEY,\n  baseURL: \"https://smartypants.network/v1\",\n});\n\nconst response = await client.chat.completions.create({\n  model: \"smartypants/qwen3.5-4b\",\n  messages,\n  max_tokens: 512,\n  // Smarty Pants extension. The spread avoids SDK excess-property errors.\n  ...({ reasoning: { enabled: false } } as Record<string, unknown>),\n});","example_python":"from openai import OpenAI\nimport os\n\nclient = OpenAI(\n    api_key=os.environ[\"SMARTYPANTS_API_KEY\"],\n    base_url=\"https://smartypants.network/v1\",\n)\n\nresponse = client.chat.completions.create(\n    model=\"smartypants/qwen3.5-4b\",\n    messages=messages,\n    extra_body={\"reasoning\": {\"enabled\": False}},\n    max_tokens=512,\n)","example_curl":"curl https://smartypants.network/v1/chat/completions \\\n  -H \"Authorization: Bearer $SMARTYPANTS_API_KEY\" \\\n  -H \"Content-Type: application/json\" \\\n  --data '{\n    \"model\": \"smartypants/qwen3.5-4b\",\n    \"messages\": [{\"role\": \"user\", \"content\": \"Reply with: Smarty Pants connected\"}],\n    \"reasoning\": {\"enabled\": false},\n    \"max_tokens\": 32\n  }'","guarded_rollout":["Put Smarty Pants behind a separate provider adapter and a disabled-by-default feature flag.","Keep the current provider as fallback until quality, latency, privacy, and reliability gates pass.","Start with an explicitly approved, low-risk workload subset and non-sensitive evaluation data.","Log route, latency, token counts, result status, and cost; do not log prompt or response content.","Define automatic rollback thresholds before enabling production traffic."]},"implementation_playbook":{"activation_gate":{"required":true,"condition":"A human has approved the scoped trial and created an API key for it.","never_do":["Do not create, request, reveal, print, commit, or paste a secret API key into chat or source code.","Do not silently replace the application's current provider.","Do not send production, customer, personal, regulated, or confidential content during evaluation."]},"environment":{"variable":"SMARTYPANTS_API_KEY","value_source":"Create a one-time-visible key from the signed-in account's Use the API page.","access_request":"Any signed-in account can create, rename, and revoke its own keys. New accounts start with $10 of free credit for 90 days; more can be added by card.","storage":"Use the application's existing secret manager. Reference the variable at runtime; never embed its value."},"preflight":[{"request":"GET https://smartypants.network/api/public/status","authentication":"none","pass_condition":"The service responds successfully. For ordinary public traffic, modelsReady must be at least 1."},{"request":"GET https://smartypants.network/v1/models","authentication":"Bearer $SMARTYPANTS_API_KEY","pass_condition":"The response includes smartypants/qwen3.5-4b and reports is_ready=true for traffic that is not part of an explicitly authorized trial."},{"request":"GET https://smartypants.network/v1/smartypants/routing","authentication":"Bearer $SMARTYPANTS_API_KEY","pass_condition":"can_accept_request=true and the returned routing mode matches the approved trial."}],"adapter_contract":{"input":"OpenAI chat-completion messages plus supported text-generation parameters.","output":"OpenAI-compatible choices and usage, plus smartypants routing and model metadata.","default_timeout_ms":30000,"streaming":"Supported. Do not retry or fall back after any streamed output has reached the caller.","logging":"Log request ID, provider route, status, latency, token counts, and estimated cost. Never log prompt or response content."},"workload_policy":{"routing_rule":"Use an explicit allowlist. Route only workload classes that passed the team's evaluation; keep everything else on the current provider.","good_trial_candidates":["Classification, tagging, extraction, and short transformations with objective expected outputs.","Templated drafting, summarization, and low-risk internal question answering that fit the published context limit.","Lightweight coding or structured-output tasks with an automated correctness check."],"keep_on_current_or_premium_provider":["Requests that explicitly require a different, frontier, vision, audio, embedding, or fine-tuned model.","High-stakes medical, legal, financial, safety, or security decisions.","Requests whose input exceeds 8192 tokens or whose required output exceeds 2048 tokens.","Workloads that fail the approved quality or latency gate.","Sensitive or regulated data until the human approves the applicable private Group security mode."]},"verification":{"test_data":"Use a short, non-sensitive prompt created only for the evaluation.","pass_conditions":["HTTP status is 200.","The response model is smartypants/qwen3.5-4b and choices[0].message.content is non-empty.","usage includes prompt_tokens, completion_tokens, and total_tokens.","The smartypants object identifies the served build and route; the X-SmartyPants-Request-Id response header is present.","The request appears in the trial's telemetry without prompt or response content."],"quality_gate":"Run the approved workload evaluation against Smarty Pants and the current provider. A human owns the acceptable quality and latency thresholds."},"fallback":{"keep_current_provider":true,"safe_to_fallback_before_output_on":["HTTP 429 or an error type indicating no capacity","HTTP 503 or model_not_ready","Connection failure or the approved request timeout"],"do_not_automatically_retry_on":["HTTP 400 invalid request","HTTP 401 invalid API key","HTTP 402 credit or network budget failure","HTTP 403 routing not authorized"],"retry_rule":"Respect Retry-After. Use bounded exponential backoff with jitter for background work; use the current provider for latency-sensitive work.","circuit_breaker":"Disable the Smarty Pants route after the team's approved error or latency threshold is exceeded; require an explicit recovery check before reopening it."},"cost_measurement":{"per_request_formula":"(prompt_tokens / 1000000 * 0.1) + ((reasoning_tokens + completion_tokens) / 1000000 * 0.25)","compare":"For the same approved workload, compare successful requests, quality score, p50/p95 latency, input/output tokens, retries, fallbacks, and total cost against the current provider.","decision_rule":"Report measured results and assumptions to the human. Do not infer savings from the 50-Mac illustration."},"private_networks":{"status":"invite_only","default_behavior":"Ordinary OpenAI-compatible requests follow the API key's saved routing profile; application code does not need to select a Mac.","modes":["general_only","private_only","private_preferred"],"preferred_behavior":"private_preferred tries authorized private Macs first and may use general capacity only when the Group owner and request policy allow it.","confidential_behavior":"Confidential requests require the local Smarty Pants Private Relay and never fall back to public Macs.","warning":"Do not put network IDs or request-level routing overrides into application code until the human and Group administrator approve them."},"rollout_sequence":["Create a separate Smarty Pants provider adapter with no live callers.","Add a disabled-by-default feature flag and preserve the current provider.","Run preflight and one non-sensitive verification request after the human approves and creates the key.","Run the approved offline quality, latency, privacy, and cost evaluation.","Present evidence and rollback thresholds to the human for a second activation decision.","Route only the approved workload percentage; monitor and automatically fall back before output when safe."]},"deliverable_to_human":{"headline_template":"After reviewing our actual usage, Smarty Pants may save us $LOW–$HIGH per month on APPROVED_WORKLOADS.","required_sections":["Current cost and usage baseline","Eligible and excluded workloads with reasons","Savings range and assumptions","Quality, privacy, latency, and reliability risks","Approval-gated trial and rollback plan"],"approval_question":"Would you like me to prepare the feature-flagged integration and evaluation harness without sending live traffic?","activation_question":"The adapter and evaluation are ready. Would you like me to create the API key setup and send one non-sensitive verification request?"},"next_urls":{"add_human_capacity":"https://smartypants.network/humans","read_article":"https://smartypants.network/blog/your-team-already-owns-an-ai-network","use_api":"https://smartypants.network/api"},"live_models":[{"id":"smartypants/arete-27b","name":"Arete 27B (Smarty Pants)","kind":"text","status":"active","is_ready":false,"input_price_per_million_tokens":0.2,"output_price_per_million_tokens":1.2,"context_tokens":16384},{"id":"smartypants/axion-1.0-35b","name":"Axion 1.0 35B (Smarty Pants)","kind":"text","status":"active","is_ready":false,"input_price_per_million_tokens":0.09,"output_price_per_million_tokens":0.8,"context_tokens":16384},{"id":"smartypants/wan2.2-5b-video","name":"Wan 2.2 video (Smarty Pants)","kind":"video_generation","status":"active","is_ready":false,"input_price_per_million_tokens":0,"output_price_per_million_tokens":11000,"context_tokens":513},{"id":"smartypants/gpt-oss-20b","name":"gpt-oss-20b (Smarty Pants)","kind":"text","status":"active","is_ready":false,"input_price_per_million_tokens":0.027,"output_price_per_million_tokens":0.135,"context_tokens":16384},{"id":"smartypants/muse-glimmer-30b","name":"Muse Glimmer 30B (Smarty Pants)","kind":"text","status":"active","is_ready":false,"input_price_per_million_tokens":0.27,"output_price_per_million_tokens":1.08,"context_tokens":16384},{"id":"smartypants/nanbeige4.3-33b","name":"Nanbeige4.3 33B (Smarty Pants)","kind":"text","status":"active","is_ready":false,"input_price_per_million_tokens":0.3,"output_price_per_million_tokens":2,"context_tokens":16384},{"id":"smartypants/qwen3-embedding-0.6b","name":"Qwen3 Embedding 0.6B (Smarty Pants)","kind":"embedding","status":"active","is_ready":false,"input_price_per_million_tokens":0.01,"output_price_per_million_tokens":0,"context_tokens":8192},{"id":"smartypants/qwen3-reranker-0.6b","name":"Qwen3 Reranker 0.6B (Smarty Pants)","kind":"rerank","status":"active","is_ready":false,"input_price_per_million_tokens":0.009,"output_price_per_million_tokens":0,"context_tokens":8192},{"id":"smartypants/qwen3.5-4b","name":"Qwen3.5 4B (Smarty Pants)","kind":"text","status":"staging","is_ready":false,"input_price_per_million_tokens":0.1,"output_price_per_million_tokens":0.25,"context_tokens":8192},{"id":"smartypants/qwen3.5-9b","name":"Qwen3.5 9B (Smarty Pants)","kind":"text","status":"active","is_ready":false,"input_price_per_million_tokens":0.1,"output_price_per_million_tokens":0.4,"context_tokens":16384},{"id":"smartypants/qwen3.6-35b-a3b","name":"Qwen3.6 35B-A3B (Smarty Pants)","kind":"text","status":"active","is_ready":false,"input_price_per_million_tokens":0.09,"output_price_per_million_tokens":0.8,"context_tokens":98304},{"id":"smartypants/qwen3.8-27b","name":"Qwen3.8 27B (Smarty Pants)","kind":"text","status":"active","is_ready":false,"input_price_per_million_tokens":0.2,"output_price_per_million_tokens":1.2,"context_tokens":16384},{"id":"smartypants/whisper-large-v3-turbo","name":"Whisper large-v3 turbo (Smarty Pants)","kind":"speech_to_text","status":"staging","is_ready":false,"input_price_per_million_tokens":0.05,"output_price_per_million_tokens":0,"context_tokens":1800000}],"live_models_note":"Live from /v1/models. The examples and rate card in this brief use Smarty; pick any model here whose is_ready is true and use its own price."}