Know what every AI your org uses and ships actually does.
Trefur traces every agent, copilot, and model call across the AI you build and the AI your team adopts — so you catch unsafe behavior before your customer does.
Two problems, one trace. The agents you ship to customers, and the copilots and agents your team adopts inside your tools — Trefur gives you a single, audit-ready record of what every one of them actually did. Traditional monitoring stops at the LLM call. Trefur picks up exactly where the interesting questions begin.
Your agent stops being a black box.
Most teams ship agents and find out what happened by reading the response. Trefur turns the agent into a live map — every node a decision, a tool call, or a retry. The pulse below is a real agent flow, simulated in your browser. In production it's your traffic, your agents, your decisions.
Every decision, every action, every outcome.
Trefur captures the full picture of agent behavior: which tools they called and why, which decisions they made at each step, where they paused, where they retried, where they failed. Not just the LLM trace — the agent trace.
Drill from a failing customer interaction down to the specific decision that went wrong, in the specific deployment, on the specific build. Find the root cause before your customer reports the bug.
Your team is already using AI you can't see.
Copilots in your IDE. Agents inside Workspace, Microsoft 365, Slack, and your CRM. Coding assistants opening pull requests. Most of it arrived without going through engineering. Trefur inventories every agent and copilot operating in your tools — including the ones you didn't know about — and gives you an audit-ready record of what each one touched.
Inventory the AI you didn't deploy
Surface unmanaged AI — the agents, copilots, and third-party AI integrations running in your tools that engineering never onboarded. See who granted access, to what data, and on whose behalf.
Catch unsafe behavior early
When an agent reaches for data it shouldn't, retries into the wrong account, or starts behaving differently after a model swap, you see it in the trace — not in a customer complaint.
Audit-ready by default
Every run is a clean, exportable record of what the AI did, when, and on whose behalf — the evidence reviewers, auditors, and your own governance team ask for.
Watch it happen — then watch it get stopped.
A support agent reads a poisoned document and tries to email your customer data to a stranger. Flip one switch, run the same attack again — this time it never leaves. No signup, no live model.
Evidence for the frameworks you report against.
AI governance is moving from policy to proof. Trefur's activity record gives you the trail that ISO 42001 and SOC 2 reviewers expect — a continuous, exportable record of how AI operates across your org. Trefur is evidence for these frameworks, not a certification claim. Each agent's trust credential works the same way: an attestation of observed fact over a stated window — “in 4,213 runs, 0 egress attempts were flagged” — not a score that certifies or guarantees future behavior.
Any agent. Any runtime. Any vendor.
Trefur is built on open standards (OpenTelemetry GenAI Semantic Conventions) so you can instrument once and capture everything — agents built on LangChain, CrewAI, AutoGen, Pydantic AI, custom MCP servers, in-house frameworks, or anything you'll be building next year.
Don't pick observability that locks you into a single agent framework. Trefur reads from all of them.
5 minutes to first trace. 30 minutes to production-ready monitoring.
Drop the SDK into your agent runtime. No re-platforming. No replatform-and-pray. Traces flow immediately; dashboards populate as the data arrives; alerts route to wherever you already operate (Slack, PagerDuty, Teams, webhook, email).
Built by engineers who have shipped agents into production and know what hurts.
- Drop-in instrumentation for 5 frameworks
- Auto-instrumentation for OpenAI, Anthropic, Bedrock, Gemini
- Native OpenTelemetry — fan out to existing backends
- Alerts to Slack, PagerDuty, webhook, email
# pip install trefur-observe
from trefur_observe import TrefurObserve, patch_langchain
from langchain.agents import AgentExecutor
obs = TrefurObserve.init(api_key="trf_obs_...")
patch_langchain(obs)
# Existing LangChain agent — no other changes
agent = AgentExecutor.from_agent_and_tools(...)
agent.invoke({"input": "Summarise this support ticket"})
# Every decision, tool call, and retry now appears in Trefur.Built for the teams shipping AI — and the teams governing it.
Engineers, platform teams, governance leads, and business operators all see different things in the same trace. Pick the closest fit.
AI engineers
Ship agents you can debug. Drop-in SDKs, the full trace, real cost numbers as you iterate.
Platform teams
One canonical trace across every team, framework, and model vendor. Fan-out friendly with what you already run.
Governance & safe AI
Inventory every agent and copilot in your tools — including the ones nobody onboarded. Audit-ready evidence, unsafe behavior caught early.
Business & ops
See what the agent did, what it cost, and where it stopped. In plain English, without a tracing tutorial.
Free for individual developers. Predictable pricing as you scale.
Start free with full feature access for personal projects. Pro tier for engineers shipping production agents. Team tier for engineering teams. Enterprise for organizations running agents at scale across multiple teams and runtimes.
See pricingTeams shipping and governing AI in production build on Trefur.
“We turned on Trefur and within a day found three AI integrations touching production data that nobody on the security team knew existed. Then we could actually trace what they were doing.”
See the AI you ship — and the AI you use.
Free tier. No card required. First trace in under five minutes.
Audit-ready traces · self-hostable collector · ISO 42001 & SOC 2 evidence