Prompt Engineering
Tool | Category | Segment | Platform / Tool | Plan / License | Monthly Price USD | Pricing Model | Free Tier / OSS | Included Usage / Limits | Prompt Assets / Versioning | Optimization / Testing | Runtime / Deployment | Integrations / Frameworks | Deployment / Hosting | Security / Privacy | Team / Governance | Best Fit | Main Limits / Caveats |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
No tagline | Prompt Engineering | Open-source prompt IDE and observability | Arize Phoenix | Open source; Phoenix Cloud also available | $0 software | Self-hosted software plus model/hosting costs; cloud service separate | ✓ | No software usage cap when self-hosted; provider API and infrastructure costs apply | Prompt Hub stores complete prompt snapshots, versions, tags, model parameters, tools and response formats | Prompt Playground compares variants, replays spans and runs prompts over datasets | Python/TypeScript clients fetch latest, exact version or environment tag; no proxy required | OpenAI and other configured providers, Phoenix clients and OpenTelemetry ecosystem | Local/self-hosted Phoenix or Phoenix Cloud | Self-hosting keeps Phoenix data in customer infrastructure; runtime provider calls still follow provider policies | Version authorship and tags support release workflows; broader access control depends on deployment | Teams wanting an OSS prompt IDE integrated with traces, datasets and experiments | Phoenix docs warn that remote prompt fetching adds a network dependency and recommend caching/fallbacks |
No tagline | Prompt Engineering | Constrained prompt programming | Guidance | MIT | $0 software | Open-source library; model and compute costs separate | ✓ | No software limits | Composable Python guidance functions interleave prompts, model generations and control flow | Regex, choice and context-free grammar constraints can be debugged with a local mock model | Runs in-process with supported local or hosted model backends | Python, Transformers/local model backends and supported API models | Local/application-hosted | Prompt logic can run locally; final data handling depends on the selected model backend | No built-in team registry or approval workflow | Developers needing precise prompt programs and token-level output constraints | Overlaps Structured Output; backend support varies and it is not a collaborative prompt CMS |
No tagline | Prompt Engineering | Automatic LLM workflow optimization | AdalFlow | MIT | $0 software | Open-source library; model/evaluation costs separate | ✓ | No software limits | PyTorch-like components represent prompts and multi-step LLM workflows as optimizable programs | LLM-AutoDiff supports zero-shot and few-shot prompt optimization using textual gradients | Optimized components run directly in Python applications and agent/RAG workflows | Python, multiple LLM providers and custom workflow components | Local/customer application infrastructure | Data path depends on selected providers; no hosted control plane is required | No built-in enterprise prompt approvals or shared hosted registry | Engineering and research teams optimizing compound RAG, chatbot and agent pipelines | Needs evaluation data and tuning expertise; broader agent framework overlap and optimization cost can be high |
No tagline | Prompt Engineering | AI gateway and prompt management | Portkey | Free cloud tier plus open-source gateway | $0 | Free tier, fixed Production plan or Enterprise; provider charges separate | ✓ | 10k recorded logs/month, 3-day log retention, 30-day metrics and 3 prompt templates | Prompt templates include variables, automatic versions, published/latest selectors and shared libraries | Multi-model playground, side-by-side comparisons, eval templates and promptfoo integration | API endpoints for saved prompts, published-version deployment, rollback and gateway routing | 1,600+ models, OpenAI-compatible SDK, LangGraph, CrewAI, promptfoo and other agent stacks | Portkey Cloud, open-source gateway, private cloud and VPC options | Privacy mode on standard plans; Enterprise adds private hosting, compliance controls and data exports | Production $49/month adds RBAC, service accounts and unlimited templates; Enterprise adds SSO and budgets | Production apps wanting prompt management and a resilient multi-provider gateway together | Free Developer plan is explicitly not intended for production; OSS gateway does not include every hosted control-plane feature |
No tagline | Prompt Engineering | Open-source LLM engineering platform | Langfuse | MIT open source plus cloud | $0 | Free cloud tier, fixed plans plus billable units, or self-hosting | ✓ | 50k units/month, 30 days data access, 2 users and all core features with limits | Prompt versioning, labels, composability, server/client caching and release management | Playground, prompt experiments, datasets and evaluation workflows | SDK prompt fetching with labels, cached fallbacks, one-click deployment and rollback patterns | Python, JavaScript, OpenTelemetry, LiteLLM, n8n and 80+ integrations | Langfuse Cloud or free self-hosting | Self-hosting keeps platform data in customer infrastructure; paid tiers add retention controls and compliance reports | Core $29/month has unlimited users; protected deployment labels and fine-grained RBAC require higher add-ons | Teams wanting a widely adopted OSS prompt registry integrated with telemetry | Prompt usage is part of a broader observability product; cloud billable-unit accounting must be modeled separately |
No tagline | Prompt Engineering | AI quality platform | Braintrust | Cloud SaaS; Enterprise deployment options | $0 | Platform fee plus processed-data and score usage | ✓ | 1 GB processed data, 10k scores, 14-day retention and unlimited users/projects/playgrounds/experiments | Prompts can be created and iterated as versioned experiment assets in projects and playgrounds | Playgrounds, experiments, datasets, scorers and Loop agent for autonomous prompt iteration | Use prompts and experiments through SDKs; environments and release controls are richer on Pro | TypeScript/Python SDKs, provider integrations and application frameworks | Hosted cloud; Enterprise offers on-prem or hosted deployment | SOC 2 Type II and MFA on Starter; Enterprise adds custom retention/export and BAA | Unlimited users on Starter; Pro is $249/month; RBAC is basic on Pro and custom on Enterprise | Engineering teams optimizing prompts against rigorous datasets and scorers | Prompt management is evaluation-centric rather than a standalone nontechnical prompt CMS |
No tagline | Prompt Engineering | Open-source prompt library and eval platform | Opik | Open source; hosted service available | $0 software | Self-hosted software plus model/hosting costs; cloud plans separate | ✓ | No software usage cap when self-hosted; infrastructure and model calls are customer costs | Project-scoped text/chat prompts, immutable sequential versions and variable templates | Prompt Playground compares variants and links exact prompt versions to traces and experiments | Python/TypeScript SDKs fetch a pinned version or latest prompt at runtime | OpenAI-style messages, Python/TypeScript SDKs and the broader Opik integration ecosystem | Self-hosted or Comet-hosted Opik | Self-hosting provides infrastructure control; cloud security follows Comet plan and terms | Projects separate prompt namespaces; hosted enterprise governance depends on plan | Teams wanting an OSS prompt library connected to experiments and traces | Prompt Library is part of a broader eval/observability system; cloud pricing was not itemized on the prompt docs |
No tagline | Prompt Engineering | Visual prompt experimentation | ChainForge | MIT | $0 software | Open-source local/web tool; provider API costs separate | ✓ | No software usage cap; public web version has a limited feature set and share limits | Visual prompt nodes, templates, variables, chains and reusable flow files | Cross-product prompt/model comparisons, code or LLM evaluators and result visualizations | Local server or limited browser app; exports flows and tabular results | OpenAI, Anthropic, Gemini, DeepSeek, Bedrock, Azure, Ollama and custom providers | Local Python server, Docker or limited chainforge.ai/play | Local deployment can keep flow data and keys under user control; provider calls follow provider policies | Flow sharing is link/file based rather than enterprise governance | Researchers and developers rapidly comparing prompt permutations across models | Latest listed release is May 2025; not a production prompt registry or runtime control plane |
No tagline | Prompt Engineering | Prompt workflow development | Microsoft Prompt flow | MIT | $0 software | Open-source SDK/CLI; Azure managed usage and model calls can add charges | ✓ | No software limits in OSS usage | Flow definitions version prompts, Python tools and model settings together in files | Local testing, batch runs, evaluations and flow comparisons support prompt iteration | Package and deploy flows to serving environments; Azure AI integrations provide managed paths | Azure AI, OpenAI, Python tools, VS Code and CI/CD workflows | Local/self-hosted or Azure-managed deployment | Local mode follows customer infrastructure; Azure security and network controls depend on selected services | Git and CI/CD provide governance; managed enterprise controls depend on Azure setup | Teams building prompt-heavy DAGs that need repeatable testing and deployment | Broader workflow framework rather than a standalone prompt registry; Azure product naming and managed features evolve |
No tagline | Prompt Engineering | Code-first prompt templates | Mirascope | MIT | $0 software | Open-source library; provider calls separate | ✓ | No software limits | Python functions and decorators define reusable prompts, messages, tools and structured outputs in code | Supports provider-agnostic experimentation and evaluation patterns but no hosted optimizer | Prompts execute directly through configured provider SDKs inside the application | OpenAI, Anthropic, Mistral, Gemini/Vertex, Groq, Cohere, LiteLLM, Azure AI and Bedrock | Local/customer application infrastructure | No Mirascope proxy is required; data goes to the selected provider | Governance follows source control, reviews and application deployment practices | Python teams wanting typed, code-native prompt composition without a separate SaaS | No visual collaborative prompt registry; overlaps general LLM application frameworks and Structured Output |
No tagline | Prompt Engineering | Textual-gradient optimization | TextGrad | MIT | $0 software | Open-source research library; optimizer and evaluator model calls cost extra | ✓ | No software limits | Treats prompts and other text variables as optimizable parameters in a computation graph | Backpropagates LLM-generated textual feedback through custom loss functions | Runs as a Python optimization loop; optimized text is exported into the application workflow | LiteLLM-supported providers, Python and PyTorch-like APIs | Local/research environment | Data is sent to configured optimization and task models; local models can change the data path | No team registry, approval workflow or managed production deployment | Researchers testing automatic prompt improvement with differentiable-programming concepts | Experimental research approach; results depend heavily on evaluator quality and repeated model calls |
No tagline | Prompt Engineering | Token-aware prompt design | Priompt | MIT | $0 software | Open-source TypeScript library; model calls separate | ✓ | No software limits; renderer performance is documented as practical around 10k scopes | JSX components define system/user messages, tools, images and prioritized context scopes | Preview tooling and source maps help inspect token budgets, ejections and cache behavior | Renders provider-ready prompt messages inside TypeScript applications | TypeScript/React-style code, Zod tools and application-specific model clients | Local/customer application infrastructure | Prompt construction stays local; rendered data follows the selected model provider | Governance is source-control based with no hosted team controls | TypeScript teams building complex prompts that must fit strict context budgets | Project describes itself as an attempt and documents priority, caching and renderer caveats; activity should be checked before critical adoption |
No tagline | Prompt Engineering | Prompt testing and optimization | promptfoo | MIT | $0 software | Open-source CLI/library; provider calls and optional hosted services separate | ✓ | No software limits; eval volume is constrained by provider budget and rate limits | Prompts live in declarative configs or files and can be parameterized across test cases | Matrix evals, assertions, red teaming and `promptfoo optimize` compare or improve prompt/provider pairs | Runs locally, in CI/CD or as part of custom hosted workflows | OpenAI, Anthropic, Azure, Bedrock, Gemini, Ollama, Portkey and many more | Local CLI, CI runner or self-hosted environment | Evals can run locally; prompts still go to configured providers unless local models are used | Git-based review and CI gates; enterprise collaboration depends on optional services | Teams treating prompts as testable code and regression-gating changes | Optimization only improves what tests measure; broad red-team/eval scope overlaps Eval Observability |
No tagline | Prompt Engineering | Open-source prompt lifecycle platform | Latitude | LGPL-3.0 open source | $0 software | Self-hosted software plus model/infrastructure costs; managed cloud availability may change | ✓ | No software usage cap stated for self-hosting | Prompt Manager supports PromptL templates, shared snippets, drafts, published versions and collaboration | Playground, datasets, evaluations, experiments and AI-generated prompt suggestions | Published prompt versions can be served through a stable AI Gateway endpoint | OpenAI, Anthropic, Gemini, Bedrock, Vercel AI SDK, LangChain, DSPy and more | Latitude Cloud or self-hosted Docker/Kubernetes-style stack | Self-hosting controls data path and provider keys; telemetry content needs explicit privacy review | Workspace collaboration; commercial licensing is available for needs beyond LGPL | Teams wanting an integrated OSS prompt editor, evaluator and gateway | Current 2026 documentation increasingly positions Latitude around agent observability, while prompt-manager docs remain available; verify product scope before adoption |
No tagline | Prompt Engineering | Discontinued prompt platform | Humanloop | Discontinued | $0; unavailable | Billing stopped July 30, 2025; service shut down September 8, 2025 | ✕ | Platform and stored data became inaccessible after the sunset date | Formerly versioned prompts, tools, flows and datasets in a central registry | Formerly supported prompt evaluation, test datasets and collaboration | No active runtime or deployment service remains | Historical SDK/API integrations only | Former hosted service; no longer available | Official migration notice required data export before shutdown | Humanloop team joined Anthropic; customers had to migrate | Migration reference and market-history comparison | Do not select for new work: official docs confirm permanent platform sunset on September 8, 2025 |
No tagline | Prompt Engineering | Automatic prompt programming | DSPy | MIT | $0 software | Open-source library; model calls and optimization runs billed by provider | ✓ | No software limits; optimization cost depends on dataset, metric and number of candidate calls | Declarative signatures and modules replace hand-maintained prompt strings with compiled programs | Optimizers tune instructions, demonstrations and sometimes weights against user-defined metrics | Compiled DSPy programs run inside Python applications with selected model providers | OpenAI, Anthropic, local models, retrieval stacks and Python ML tooling | Runs locally or in customer application infrastructure | Data goes to whichever model/evaluation providers are configured | No built-in SaaS governance; version compiled artifacts and datasets in existing engineering systems | Teams with measurable tasks that want systematic optimization instead of manual prompt edits | Requires representative train/dev data and metrics; optimization can consume many model calls and may overfit |
No tagline | Prompt Engineering | Prompt-learning research framework | OpenPrompt | Apache-2.0 | $0 software | Open-source research framework; training/inference compute separate | ✓ | No software limits | Templates and verbalizers define prompt-learning pipelines over pretrained language models | Supports manual and learned templates, verbalizers and optimization strategies for prompt learning | Runs training and inference through PyTorch and Hugging Face model pipelines | PyTorch and Hugging Face Transformers | Local/research compute | Data stays in the configured training environment unless external services are added | No hosted team governance or production release workflow | NLP researchers studying prompt learning and parameter-efficient task adaptation | Latest listed release is April 2022 and docs acknowledge outdated sections; not designed for modern hosted chat prompt management |