Respan

Respan is the full-stack AI engineering platform for LLM and agent products. It helps you route models, set up monitoring, build prompts, run evaluations, and audit agent security, all in one place.

To get started, create a free account at platform.respan.ai.

Respan in 6 Minutes: Gateway, Observability, Evals, Prompts, Red Teaming

How Respan works

Everything in Respan is built on one core data structure: the span. Whether it is a single LLM call, a full agent workflow, or a tool step like retrieval (RAG), a memory lookup, an embedding, or a reranking, every interaction is stored as a span with its input, output, model, metrics, and metadata.

Every product reads and writes this same span data:

• Gateway: route any LLM call through one endpoint and log it as a span.
• Observability: trace agent workflows as a span tree, and monitor live metrics with alerts.
• Prompt management: version-control your prompts and serve them directly through the gateway.
• Evaluations: score a single span or a whole trace against your evaluators.
• Red Team: run an authorized security audit of your agent.

Gateway

Route every LLM request through a single endpoint to 1000+ models, with automatic logging, fallbacks, retries, load balancing, and caching. Point any LLM SDK at Respan, change the base URL, and the rest of your code stays the same.

Set up the gateway · Model catalog · Reliability & caching

Observability

Set up tracing and monitoring for your LLM traffic: inspect every span, watch live metrics, catch error incidents, and read the behavioral signals behind them.

Set up the SDK · Framework integrations · OpenTelemetry

Prompt management

Ship prompts without shipping code: create templates with {{variables}}, commit and compare versions, test them in the playground, and deploy instantly. Your application picks up the new version the moment you publish.

Create prompts · Deploy & version

Evaluations

Measure output quality and catch regressions before they reach users: grade with LLM judges, deterministic code checks, or human review, across curated datasets, offline experiments, and live traffic.

1. Build a dataset: sample production spans or import test cases.
2. Set up evaluators: LLM judges, code checks, or human review.
3. Run experiments: test against different prompt versions or models, then compare scores.
4. Add online evals: score live traffic and alert when quality drops.

Red Team

Run an authorized security audit of your agent: adversarial campaigns that probe for prompt injection, system-prompt leakage, secret disclosure, and goal hijacking, then report confirmed findings with severity. It runs against the same connected agent you already trace.

Only test an agent and its connected systems when you own them or have explicit authorization to assess them.

Run a campaign