Integration guides
26 frameworks × 4 capabilities. Each guide is the exact diff for that framework, the switch that turns the capability on, and what it saves on the framework's usual model.
104 guides cover 26 frameworks in Python, TypeScript, YAML and shell, each with the real base-URL change and one of 4 capabilities: semantic caching, cascade routing, prompt compression and token-budget guardrails. Prices come from table 2026-10-26; gateway overhead is 1.52 ms p50 measured, and the free tier covers $500 of routed spend a month.
· Source: pricing table 2026-10-26; the diffs are the SDKs' documented configuration options
At a glance
- Guides
- 104 (computed)
- Frameworks
- 26
- Capabilities
- semantic caching, cascade routing, prompt compression, token-budget guardrails
- Languages
- TypeScript, Python, Settings, Shell, YAML
- Request shapes
- OpenAI chat / Responses, Anthropic Messages, Gemini native
- Gateway overhead
- 1.52 ms p50 (measured)
- Parity floor
- 99.8%
- Free tier
- $500/month routed
Capabilities and how each is switched on
| Capability | Per request | Per workspace | Watch |
|---|---|---|---|
| semantic caching | X-TokenZero-Semantic-Cache: true (exact cache is automatic for temperature-0 or seeded requests; X-TokenZero-Cache: always forces it) | semantic_cache: true, cache_ttl: 3600 | cache_hit_rate_pct and savings from engine cache on the scorecard |
| cascade routing | X-TokenZero-Cascade: auto (default advise: the recommendation is returned in X-TokenZero-Cascade-Recommendation and nothing moves) | cascade: auto, shadowEval: true | routed share and parity per pair on the scorecard |
| prompt compression | X-TokenZero-Cache-Strategy: static (default) · X-TokenZero-Prompt-Cache-TTL: 1h or 12h | cache_strategy: static, prompt_cache_ttl: 1h | input_tokens_cached_pct and compaction tokens saved on the scorecard |
| token-budget guardrails | X-TokenZero-Budget-Usd / -Spent-Usd / -Remaining-Usd / -Resets-At on every response | monthly_budget_usd: 500 | spend against budget on the dashboard |
Agent and orchestration frameworks (40)
Next.js + Vercel AI SDK: semantic cachingNext.js + Vercel AI SDK: cascade routingNext.js + Vercel AI SDK: prompt compressionNext.js + Vercel AI SDK: token-budget guardrailsLangChain (Python): semantic cachingLangChain (Python): cascade routingLangChain (Python): prompt compressionLangChain (Python): token-budget guardrailsLangChain.js: semantic cachingLangChain.js: cascade routingLangChain.js: prompt compressionLangChain.js: token-budget guardrailsLlamaIndex (Python): semantic cachingLlamaIndex (Python): cascade routingLlamaIndex (Python): prompt compressionLlamaIndex (Python): token-budget guardrailsPydantic AI: semantic cachingPydantic AI: cascade routingPydantic AI: prompt compressionPydantic AI: token-budget guardrailsAutoGen (AgentChat): semantic cachingAutoGen (AgentChat): cascade routingAutoGen (AgentChat): prompt compressionAutoGen (AgentChat): token-budget guardrailsCrewAI: semantic cachingCrewAI: cascade routingCrewAI: prompt compressionCrewAI: token-budget guardrailsDSPy: semantic cachingDSPy: cascade routingDSPy: prompt compressionDSPy: token-budget guardrailsSemantic Kernel (Python): semantic cachingSemantic Kernel (Python): cascade routingSemantic Kernel (Python): prompt compressionSemantic Kernel (Python): token-budget guardrailsHaystack: semantic cachingHaystack: cascade routingHaystack: prompt compressionHaystack: token-budget guardrails
Provider SDKs (24)
Vercel AI SDK Core (Node): semantic cachingVercel AI SDK Core (Node): cascade routingVercel AI SDK Core (Node): prompt compressionVercel AI SDK Core (Node): token-budget guardrailsOpenAI Node SDK: semantic cachingOpenAI Node SDK: cascade routingOpenAI Node SDK: prompt compressionOpenAI Node SDK: token-budget guardrailsOpenAI Python SDK: semantic cachingOpenAI Python SDK: cascade routingOpenAI Python SDK: prompt compressionOpenAI Python SDK: token-budget guardrailsAnthropic Python SDK: semantic cachingAnthropic Python SDK: cascade routingAnthropic Python SDK: prompt compressionAnthropic Python SDK: token-budget guardrailsAnthropic TypeScript SDK: semantic cachingAnthropic TypeScript SDK: cascade routingAnthropic TypeScript SDK: prompt compressionAnthropic TypeScript SDK: token-budget guardrailsGoogle GenAI SDK: semantic cachingGoogle GenAI SDK: cascade routingGoogle GenAI SDK: prompt compressionGoogle GenAI SDK: token-budget guardrails
Coding agents and IDEs (16)
Cursor: semantic cachingCursor: cascade routingCursor: prompt compressionCursor: token-budget guardrailsClaude Code: semantic cachingClaude Code: cascade routingClaude Code: prompt compressionClaude Code: token-budget guardrailsOpenAI Codex CLI: semantic cachingOpenAI Codex CLI: cascade routingOpenAI Codex CLI: prompt compressionOpenAI Codex CLI: token-budget guardrailsCline: semantic cachingCline: cascade routingCline: prompt compressionCline: token-budget guardrails
Proxies (4)
LiteLLM Proxy: semantic cachingLiteLLM Proxy: cascade routingLiteLLM Proxy: prompt compressionLiteLLM Proxy: token-budget guardrails
Runtimes and deployment (20)
FastAPI service: semantic cachingFastAPI service: cascade routingFastAPI service: prompt compressionFastAPI service: token-budget guardrailsCloudflare Workers: semantic cachingCloudflare Workers: cascade routingCloudflare Workers: prompt compressionCloudflare Workers: token-budget guardrailsSupabase Edge Functions: semantic cachingSupabase Edge Functions: cascade routingSupabase Edge Functions: prompt compressionSupabase Edge Functions: token-budget guardrailsAWS Lambda: semantic cachingAWS Lambda: cascade routingAWS Lambda: prompt compressionAWS Lambda: token-budget guardrailsDocker Compose: semantic cachingDocker Compose: cascade routingDocker Compose: prompt compressionDocker Compose: token-budget guardrails
Questions people ask
Do I need to change my prompts or models?
No. Every guide is a base-URL or environment change; prompts, models, tools and keys stay as they are. Capabilities are switched on per request with a header or per workspace with a setting.
Which frameworks are covered?
Next.js + Vercel AI SDK, Vercel AI SDK Core (Node), LangChain (Python), LangChain.js, LlamaIndex (Python), Pydantic AI, OpenAI Node SDK, OpenAI Python SDK, Anthropic Python SDK, Anthropic TypeScript SDK, Google GenAI SDK, Cursor, Claude Code, OpenAI Codex CLI, Cline, AutoGen (AgentChat), CrewAI, DSPy, Semantic Kernel (Python), Haystack, LiteLLM Proxy, FastAPI service, Cloudflare Workers, Supabase Edge Functions, AWS Lambda, Docker Compose.
How are the savings on each guide computed?
From pricing table 2026-10-26 on a reference call of 2,000 input (1,500 static) and 300 output tokens at 100,000 calls a month, on the model the framework most commonly defaults to. Your own numbers come from the calculator and, once live, from signed receipts.