---
title: "Caveman switching"
description: "Move an optimization workflow to Caveman, or add it beside your current tools. Each guide covers setup, validation, and rollback."
canonical: https://caveman.so/switch
last-updated: 2026-09-07
---

# Caveman switching

Move an optimization workflow to Caveman, or add it beside your current tools. Each guide covers setup, validation, and rollback.

- [Test Caveman alongside AgentOps in a multi-agent workflow](https://caveman.so/switch/agentops): Preserve AgentOps session and agent identities while evaluating Caveman. Check handoffs, recovery access, duplicate instrumentation, and full-task accounting.
- [Add Caveman to an Arize Phoenix workflow](https://caveman.so/switch/arize-phoenix): Preserve Phoenix traces and evaluations while testing Caveman compression. Map OpenTelemetry fields and validate a private Platform pilot before migration.
- [Evaluate Caveman while keeping your Bifrost gateway](https://caveman.so/switch/bifrost): Preserve Bifrost provider mappings, virtual keys, plugins, and fallbacks while testing Caveman context reduction. Validate one caller and keep a direct rollback.
- [Test Caveman with your Braintrust evaluation baseline](https://caveman.so/switch/braintrust): Move a Braintrust optimization workflow carefully: preserve datasets and scorers, test Caveman as a candidate, and validate Platform ingestion before cutover.
- [Evaluate Caveman from a Cloudflare AI Gateway stack](https://caveman.so/switch/cloudflare-ai-gateway): Move one AI workload into a local Caveman trial. Keep Workers and gateway policies explicit, validate cache behavior, and preserve rollback.
- [Switch a Headroom workflow to Caveman](https://caveman.so/switch/headroom): Replace one Headroom compression path with Caveman. Preserve recovery data, avoid port conflicts, validate full tasks, and keep rollback simple.
- [Add Caveman to a Helicone workflow](https://caveman.so/switch/helicone): Test Caveman compression while keeping Helicone visibility. Preserve session identifiers, compare complete runs, and plan any telemetry move explicitly.
- [Move an agent optimization workflow from Langfuse to Caveman](https://caveman.so/switch/langfuse): Preserve Langfuse traces, prompts, and evaluations while testing Caveman. Map event fields, check capture policies, and validate a reversible Platform pilot.
- [Add Caveman observations to LangGraph.js](https://caveman.so/switch/langgraph): Attach the experimental Caveman LangGraph.js adapter while preserving graph state, callbacks, checkpoints, and retries. Validate usage identity and rollback.
- [Evaluate Caveman beside LangSmith without breaking agent workflows](https://caveman.so/switch/langsmith): Plan a LangSmith migration by preserving trace links, datasets, prompt references, and LangGraph state. Test Caveman locally or through a scoped Platform pilot.
- [Add Caveman to LiteLLM: setup, checks and rollback](https://caveman.so/switch/litellm): Keep LiteLLM keys and routing while testing Caveman compression. Configure a local deployment or agent path, verify recovery, and roll back.
- [Move a LLMLingua compression step to Caveman](https://caveman.so/switch/llmlingua): Test Caveman against an existing LLMLingua pipeline with saved inputs, byte-exact recovery checks, latency measurement, and rollback.
- [Evaluate Caveman from a Martian gateway or router](https://caveman.so/switch/martian): Map Martian aliases and provider credentials, test Caveman on one workload, and preserve a working gateway or routing baseline.
- [Add Caveman to a Mastra agent with native processors](https://caveman.so/switch/mastra): Integrate the experimental Caveman Mastra processor from source. Preserve processor order, inspect raw usage, test streams and cache hits, and keep rollback simple.
- [Evaluate Caveman alongside Not Diamond routing](https://caveman.so/switch/not-diamond): Compare a Caveman routing pilot with an existing Not Diamond policy. Preserve model pools, separate prompt changes, and keep a working fallback.
- [Move an OpenRouter caller to a Caveman local trial](https://caveman.so/switch/openrouter): Test a direct-provider request through Caveman while preserving OpenRouter configuration. Map model IDs, credentials, capabilities, and rollback.
- [Test Caveman beside a Portkey gateway](https://caveman.so/switch/portkey): Add local Caveman compression to a Portkey workflow, or test one direct-provider caller. Preserve guardrails, virtual-key policy, and rollback.
- [Add Caveman without breaking PromptLayer prompt releases](https://caveman.so/switch/promptlayer): Plan a PromptLayer migration around prompt versions, variables, workflow dependencies, and evaluation checks. Test Caveman context reduction before moving production flows.
- [Test Caveman context compression with a Pydantic AI application](https://caveman.so/switch/pydanticai): Keep Pydantic AI execution and validation while testing Caveman on tool output. Define recovery, check provider boundaries, and plan a full SDK migration only when needed.
- [Move a RouteLLM experiment into a Caveman pilot](https://caveman.so/switch/routellm): Preserve a RouteLLM baseline, model pair, and calibration set while evaluating Caveman. Check capability limits, full-task cost, and rollback.
- [Switch an RTK command workflow to Caveman](https://caveman.so/switch/rtk): Compare RTK with Caveman shrink or a native agent wrapper. Preserve hooks, check exit codes and original output, then roll out one workflow.
- [Evaluate Caveman from a Vercel AI Gateway setup](https://caveman.so/switch/vercel-ai-gateway): Test Caveman without rewriting your Vercel app. Preserve model and auth settings, distinguish AI SDK adapters from gateway changes, and roll back.
- [Add Caveman to a Vercel AI SDK agent](https://caveman.so/switch/vercel-ai-sdk): Attach the experimental Caveman adapter to an existing AI SDK loop. Build from source, preserve callbacks, test streaming and usage, and remove it cleanly.