Feature  ·  Playground

Find prompt failures in the sandbox, not in production.

Test prompts across 4 providers with real streaming, tool calls, structured output, and your actual guardrails — before any code ships.

OpenAI, Anthropic, Google, OpenRouter. Tool calls. Structured output. Guardrail integration.

Start free →View docs
Zespan Playground
Works withOpenAIAnthropicGoogle GenAIOpenRouterTool callsStructured output

4

providers

100+

models

Live

streaming

1.0  Playground

4 Providers, 100+ Models

What you get

OpenAI: GPT-4o, GPT-4o-mini, o1, o3-mini, GPT-4-turboAnthropic: Claude 3.5 Sonnet, Claude 3 Opus, Claude 3 HaikuGoogle: Gemini 1.5 Pro, Gemini 1.5 Flash and all Google GenAI modelsOpenRouter: 100+ models (Llama 3, Mixtral, Yi, DeepSeek, and more)

OpenAI (GPT-4o, GPT-4o-mini, o1, o3), Anthropic (Claude 3.5 Sonnet, Claude 3 Opus, Haiku), Google (Gemini 1.5 Pro, Gemini 1.5 Flash), and OpenRouter (100+ models). Available models are fetched dynamically — always current.

2.0  Playground

Tool Calls & Structured Output

What you get

Tool definitions: pass function schemas, see tool call arguments and resultsStructured output: pass a JSON schema — see if the model compliesMultiple tool calls: test models that invoke multiple tools in sequence

Pass tool/function schemas to test tool-calling before wiring up real integrations. Pass a JSON schema to enforce structured output and validate compliance immediately. Catch schema mismatches and tool argument errors before they reach production.

3.0  Playground

Streaming & Config Overrides

What you get

Real-time token streaming — same experience as productionConfig overrides: temperature, max_tokens, top_p, and provider paramsText mode and Chat mode (system/user/assistant/tool message array)

Stream completions token by token — identical to production streaming behavior. Override temperature, max_tokens, top_p, and any provider-specific parameter to fine-tune behavior in the sandbox.

4.0  Playground

Guardrail Integration

What you get

applyGuardrails: true — applies all project guardrails to playground runsSee block/warn/redact behavior before any prompt reaches productionTest guardrail rules against new prompts without a live request

Apply your project's guardrails to Playground runs. The same PII, toxicity, topic boundary, and custom rules that run in production run in the sandbox. Test prompt safety interactively before deploying.

Setup

Under 5 minutes,
two lines of code.

No forking and no architecture changes. Traces appear within seconds of the first agent run, with cost attribution, eval scores, and anomaly alerts on by default.

typescript
// Playground is in-product — no SDK setup required.
// Access it from the sidebar: Playground.

// What you can test:
// - Text mode: single string prompt
// - Chat mode: multi-turn message array (system / user / assistant / tool)
// - Tool definitions: pass function schemas to test tool-calling
// - JSON schema output: validate structured output compliance
// - Guardrails: apply project guardrails to sandbox runs

Common questions

Do Playground runs appear in my trace data?

Yes. Playground runs are traced like any other LLM call. You can find them in the Trace Explorer filtered by environment=playground or operation=playground-run.

Do I need API keys for each provider?

Yes. Each provider (OpenAI, Anthropic, Google) requires its own API key, which you configure in Project Settings → Providers. Zespan doesn't proxy through its own API keys for providers.

What's the difference between Chat mode and Text mode?

Text mode is a single string prompt — equivalent to a completion or a system prompt. Chat mode is a multi-turn message array with system, user, assistant, and tool roles — equivalent to the chat completions API. Use chat mode to test multi-turn conversations and system prompt behavior.

Can I test a prompt in the Playground before promoting it to production?

Yes, and this is the intended workflow. Load the prompt version from Prompt Management into the Playground, test it with guardrails enabled, and if it passes, promote it to the production label. The Playground is your manual safety check; Simulations are your automated check.

Explore more features

All features →
TracingSee exactly what your agents are doing.Every LLM call, agent step, and tool invocation captured as a structured trace — with cost, latency, and tokens per span.Agent MonitoringKnow which agents are healthy, and which aren't.Composite health scores, delegation graphs, and per-agent cost attribution — built for systems with many cooperating AI agents.EvaluationsMeasure output quality on every trace, automatically.12 built-in LLM-as-judge templates run on every new trace with no setup. Track quality trends, catch regressions, and run manual eval campaigns.GuardrailsStop bad outputs before they reach users.7 guardrail types run inline on every LLM request — block, warn, redact, or log. PII, toxicity, topic drift, format, cost ceiling, and custom rules.

Your agents are running.
Do you know what they’re doing?

Observe, evaluate, guard, and control every agent, live in minutes. No credit card required.

Start free →Talk to us →
Free tier availableUnder 5 min setupNo credit card