# Bento — Self-learning production infrastructure for AI agents

> The closed-loop platform for teams to ship AI agents that compound. Monitor what runs. Improve what fails. Evaluate what changes. OpenTelemetry-native across every framework.

Built by BentoLabs (YC P26). [Book a demo](https://cal.com/team/bentolabs-ai/platform-demo).

## The problem

Building AI agents is easier than ever. Trusting one in production isn't.

Prompt changes, model swaps, and new tools reach production as silent regressions — caught by your customers, not your dashboards. When you do find a regression, the context is scattered: traces in one tool, evals in another, fixes in a third. None of it feeds back into the agent.

So every release starts from zero. Learnings don't compound. Neither does your AI investment.

## The solution

Bento is the closed-loop platform for teams to ship self-improving agentic systems. One stack covers the full lifecycle of a production agent — observation, improvement, evaluation, and the feedback loop between them.

### Monitor

- **Every span, captured.** OpenTelemetry-native traces from every framework you run. Jump into the span tree, or follow a signal badge straight to the call that broke.
- **Regression signals.** Describe any failure mode in plain English. Bento trains a signal on your own traces, fires it in real time, and backfills your entire history to show how long it's been happening.
- **Alerts in plain English.** Bento groups fires into incidents and judges each one real or benign — so you only wake up for real drift.
- **Behavioural drift.** See when your agent starts behaving differently. Bento pinpoints what shifted.

### Improve — every production run makes the agent better

- **Artifacts.** Reusable, trigger-based fixes — skills, subagents, and tools that graduate from candidate to promoted as they prove themselves in production.
- **The Book.** A living memory of what your agent has learned: every failure pattern, fix, and outcome, in plain language.
- **Evaluations.** Every release scored against your production history — offline, in CI, and on live traffic. Catches regressions before users do.
- **Versions.** Every prompt, skill, and model change is versioned, diffable, and reversible. Trace any regression back to the change that caused it.

## How Bento compares

- **vs. LangSmith / Langfuse.** Trace tools tell you what happened. Bento closes the loop — it reads production traces, distils failure patterns into reusable skills, and injects them into the next run.
- **vs. Datadog / New Relic.** APMs speak HTTP and databases. Bento is OpenTelemetry-native but built around agent semantics: cost per run, tool failure rates, drift over prompt versions.
- **vs. DSPy / OPRO.** Prompt optimisers tune wording. Bento improves full agent behaviour — tool selection, routing, mid-run interventions. Prompt wording is roughly 20% of why agents fail in production.
- **vs. fine-tuning / RL.** Bento's output is readable Markdown and YAML in your repo, not opaque weights. Skills are model-agnostic; they transfer when you upgrade models.

## Who it's for

Teams running AI agents in production — coding agents, customer-support agents, vertical agents in payments, healthcare, compliance, and operations — that need their agents to improve over time, not just stay alive.

## Get started

- [Book a demo](https://cal.com/team/bentolabs-ai/platform-demo)
- [Read the blog](https://bentolabs.ai/blog)
- [FAQs](https://bentolabs.ai/faqs)
- [Contact the team](https://bentolabs.ai/contact)