Reliable AI agents.
Repeatable outcomes.

Jetty makes AI agents reliable and repeatable. They check their own work and improve until it’s right. Run a task once or a thousand times, get consistent results.

Describe a task you need done reliably and repeatably.

Using a coding agent? Install the Jetty skill.

Trusted by teams at

Brickroad AI Akinox

How it works

From instruction to results in one API call.

Write a runbook. Call the API. The agent handles the entire pipeline autonomously — the runbook is your contract, the agent figures out the implementation.

runbook.md — ETL Pipeline Agent

---

version: "1.0.0"

evaluation: rubric

agent: claude-code

snapshot: python312-uv

---

# ETL Pipeline Agent

## Objective

Fetch new events, enrich with

summaries, persist to output.

## Parameters

| source | {{source_table}} | raw.events |

## Step 1: Fetch Records

Query source_table for new rows...

## Step 2: Enrich & Transform

Apply Summarizer skill to each...

## Evaluation

| Completeness | 5: all | 1: gaps |

Pass ≥ 4.0. Max 3 retries.

Jetty Agent Execution Trace

etl-pipeline-agent v1.0.0

📋

Runbook loaded

Parsed 3 steps, 2 dependencies

✓ done

🗄️

db:query_skill

Fetching 2,847 new events

✓ done

🧠

summarizer_skill

Processing batch 12 of 29...

running

💾

write_results_skill

Awaiting enriched records

waiting

⚖️

rubric_eval_skill

Quality evaluation pending

waiting

1.Write a runbook

A markdown document that becomes the agent’s mission. Use the Jetty skill in your editor to help you write it — or start from a template. Not a prompt — a full spec with steps, tools, and evaluation criteria.

Jetty SkillMarkdownVersion control

2.Call the API

Send your runbook through the Jetty completions endpoint. Jetty provisions an isolated sandbox, installs the agent, and it executes freely — shell, network, Python, browser automation. Full autonomy.

Jetty APISandboxedFull autonomy

3.Get results back

Output files, execution trajectory, and real-time progress via streaming or webhook. Every artifact persisted to cloud storage.

Jetty PlatformStreamingWebhooksCloud storage

Supercharge your agent with the Jetty skill

Multi-step workflow, LLM judges, monitoring and more

Runbooks

A runbook is a recipe for your AI agent.

Write once, run anywhere. Runbooks encode your agent’s objective, parameters, dependencies, and step-by-step logic in a single, portable Markdown file.

→

YAML frontmatter

Version, evaluation strategy (programmatic or rubric), agent, model, and snapshot environment.

→

Objective and output manifest

What the agent is doing and the exact files it must produce. The task isn’t complete until every file exists.

→

Parameters and dependencies

Template variables injected at runtime, plus tools and skills the agent needs. The runtime checks availability before execution.

→

Steps

Sequential plain-language instructions. Each step can run code, call tools, or invoke skills.

→

Evaluation and iteration

Rubric-based scoring or programmatic validation. If the agent fails, it retries with bounded iteration (typically 3 rounds).

Learn how to write runbooks →

RUNBOOK-etl-pipeline.md

---

version: "1.0.0"

evaluation: rubric

agent: claude-code

model: claude-sonnet-4-6

snapshot: python312-uv

---

# ETL Pipeline Agent

## Objective

Fetch new events from the source database,

enrich each record with AI-generated summaries,

and persist the results to the output table.

Produce a validation report and summary.

## REQUIRED OUTPUT FILES

You MUST write all of the following files.

The task is NOT complete until every file

exists and is non-empty.

| {{results_dir}}/validation_report.json |

| {{results_dir}}/summary.md |

| {{results_dir}}/enriched_events.csv |

## Parameters

| Source table | {{source_table}} | raw.events | Input |

| Results dir | {{results_dir}} | /app/results | Output |

| Batch size | {{batch_size}} | 100 | Records per batch |

## Dependencies

| Database | Tool | Yes |

| Summarizer | Skill | Yes |

| CSV Writer | Tool | Yes |

## Step 1: Environment Setup

Create {{results_dir}} if it doesn't exist.

Verify database connectivity and

Summarizer skill availability.

## Step 2: Fetch Records

Query {{source_table}} for new rows since

last checkpoint. Process in batches of

{{batch_size}}.

## Step 3: Enrich & Transform

For each batch, invoke the Summarizer

skill to generate a summary for each

event record. Append to enriched dataset.

## Step 4: Write Results

Persist enriched records to

{{results_dir}}/enriched_events.csv.

## Evaluation

Pass if score ≥ 4.0, no criterion below 3.

## Iteration

If evaluation fails, review the Common

Fixes table and retry. Max 3 rounds.

| Failure | Fix |

| Missing rows | Re-fetch with offset |

| Bad summaries | Increase context window |

| Schema errors | Validate before write |

## Final Checklist

```bash

#!/bin/bash

for f in validation_report.json summary.md

enriched_events.csv; do

[ -s "{{results_dir}}/$f" ] || exit 1

done

```

You can’t improve what you can’t see.

You can’t trust what you can’t trace.

Every Jetty run ships with a full execution trace, rubric-based evaluation scores, and output artifacts. Agents don’t just run — they self-evaluate against your quality criteria and iterate until they pass.

Ready to get started?

Give your agent its first mission.

Describe the job. Jetty turns it into a reusable tool you can run any time.

Get started

Backed by

Hidden Layers

Reliable AI agents.Repeatable outcomes.