CHAPTER 04.2 — JSOC AI · PROOFSTATE
PROOFSTATE
CONFIDENCE IN EVERY AI RELEASE.
- 01WHAT IT IS
- One workspace to trace, evaluate and improve AI applications — from the first trace to the next release.
- 02WHO IT IS FOR
- Teams shipping AI to real users: engineers, product owners and the reviewers who sign off.
- 03WHAT YOU GET
- A record of every step, a score behind every change, and evidence behind every release.
/ THE REAL PRODUCT
SEE THE
EVIDENCE.
Real screens · demo data · open any of them
/ THE IMPROVEMENT LOOP
OBSERVE. EVALUATE.
IMPROVE. PROTECT.
Four moves, on repeat. Pick one, or let it play.
/ 01 — OBSERVE
FIND THE SIGNAL.
See every step your AI took — the prompt, the tools it called, how long it took and what it cost.
- Traces of every request, span by span
- Sessions, users and an agent graph
- Latency and cost per step
/ EVERYTHING IN THE WORKSPACE
TWELVE TOOLS.
ONE PLACE.
- TRACE EXPLORER
Follow any request from first token to final answer.
- AGENT GRAPH
See how multi-step agents actually move through tools.
- SESSIONS & USERS
Group traces by conversation and by user.
- ANALYTICS
Quality, latency and cost over time.
- ALERT RULES
Hear about drift before your customers do.
- PROMPT LIBRARY & STUDIO
Versioned, labelled prompts and a safe place to iterate.
- SCORES & HUMAN REVIEW
Reviewers and judges on the same scorecard.
- DATASETS & EXPERIMENTS
Real cases, repeatable tests, side-by-side results.
- RED TEAMING
Adversarial probes, kept as release evidence.
- VERA, THE ASSISTANT
Ask about your traces in plain language.
- AI GATEWAY
One door to your model providers.
- AUTOMATIONS & INTEGRATIONS
Wire ProofState into the rest of your stack.
/ INTEGRATIONS
KEEP YOUR STACK.
ADD THE EVIDENCE.
ProofState plugs into the providers and frameworks you already use. OpenTelemetry native.
MODEL PROVIDERS
- OpenAI
- Anthropic
- Gemini
- Vertex AI
- Amazon Bedrock
- Mistral
- Groq
- Ollama
- OpenRouter
- Together AI
AGENT FRAMEWORKS & SDKS
- LangChain
- LlamaIndex
- CrewAI
- Mastra
- Pydantic AI
- Vercel AI SDK
- Hugging Face
- + DSPY · HAYSTACK · AGNO · LITELLM · PORTKEY · GUARDRAILS
STANDARDS & PROTOCOLS
- OpenTelemetry
- MCP
From ProofState's integration catalog. Brand names belong to their owners.
/ GET STARTED
FOUR STEPS
TO YOUR FIRST TRACE.
- 01CREATE A PROJECT
Sign in at proofstate.ai and create a project for your application.
- 02ADD AN LLM CONNECTION
Under Project Settings, connect a model provider for the Playground and experiments.
- 03CREATE AN API KEY
Save the secret when it is shown — it is displayed only once.
- 04SET THREE VARIABLES
Point the SDK at your project, run your app, and watch the traces arrive.
PROOFSTATE_PUBLIC_KEY="pk-ps-..."
PROOFSTATE_SECRET_KEY="sk-ps-..."
PROOFSTATE_BASE_URL="https://proofstate.ai"SDKs for JavaScript, Python and Java · public API · project MCP server
/ FREQUENTLY ASKED QUESTIONS
QUESTIONS
ABOUT PROOFSTATE.
HOW DO WE SEND DATA TO PROOFSTATE?
With the SDKs for JavaScript, Python or Java, the public API, or OpenTelemetry. Create a project, add a project API key, and set three environment variables in your application.
DOES IT WORK WITH OUR FRAMEWORK?
Most likely. ProofState integrates with the major model providers and frameworks — OpenAI, Anthropic, Google, Amazon Bedrock, Mistral, LangChain, LlamaIndex, CrewAI, Mastra and more — and anything that speaks OpenTelemetry.
CAN OUR TEAM USE IT FROM THEIR AI TOOLS?
Yes. Each project has an MCP server, so assistants and agents your team already uses can query traces, prompts and scores directly.
HOW DOES IT RELATE TO FABRIC?
Fabric sends every governed run to ProofState as evidence, and can use prompts kept in ProofState — always the version labelled production. Each product also works on its own.
YOUR NEXT STEP
SEE YOUR OWN TRACES IN PROOFSTATE.
Bring one AI feature. We'll help you instrument it, build a first test set, and show you what the evidence looks like.
