Jetty

Each workload runs in a sandbox with a coding agent, a model and the integrations it needs, with one integration on the free plan and ten on Squadron Platform, plus webhooks. Integrations are credentials injected into every run as environment variables for Slack, databases, GitHub, CRMs or the customer's own app, which the agent then uses through its own tools. There is no native connector catalog with defined read or write actions.

SourceJetty, jetty.io pricing, homepage and llms.txtread 2026-09-21

Workflows chain prebuilt actions in an ordered list of steps, and each action does one piece of work with no agent in the loop. Path expressions wire the steps together by reading earlier steps' outputs, and control flow actions include a fan out that starts a child run for each item in a list. A runbook hands a whole job to an agent in the same system, so the platform carries the control flow across steps and mixes deterministic steps with agent steps.

SourceJetty, jetty.io/docs/reference/api/stepsread 2026-09-21

A runbook carries the job and its checks, and integrations pass credentials for the customer's systems into each run, so knowledge reaches the agent in each run through its own tools. There is no document ingestion, index or retrieval store over the customer's knowledge.

SourceJetty, jetty.io/docs/get-started and llms.txtread 2026-09-21

People decide which tasks become workloads, investigate finished runs and edit the runbook for the next run. Jetty lists adding guardrails as a way to improve a workload but describes no mechanism. There is no approval step, consent checkpoint, escalation rule or pause for a person's sign off during a run.

SourceJetty, jetty.io homepage and docs/get-startedread 2026-09-21

Squadron Platform adds roles and permissions and unlimited API keys, and API credentials are obtained and revoked through OAuth. The volume plan lists custom SLA, SOC 2 and HIPAA and single tenant options as a contractual offering, with no report, auditor or date published, and Jetty links no trust center.

SourceJetty, jetty.io/pricing and llms.txtread 2026-09-21

Jetty tracks everything the agent does in its managed sandbox. It records every step and the outputs of each run, with a workload report on how well the agent performed, and surfaces costs and errors for investigation. Each run is captured end to end, with every step's inputs, outputs, files and status.

SourceJetty, jetty.io/docs/get-started, homepage and llms.txtread 2026-09-21

Every run's configuration and trajectory is stored, and what runs reveal feeds edits to the runbook. That is a record of runs and of improvements absorbed into the runbook, not a memory layer the agent reads across runs, and there is no memory with a stated scope or lifetime.

SourceJetty, jetty.io homepage and docs/get-startedread 2026-09-21

Standard plans run workloads in Jetty's managed sandboxes. The volume plan offers single tenant options, and Jetty invites customers to discuss flexibility on data residency and deployment, including single tenant cloud. There are no named regions, customer environment option or region selection.

SourceJetty, jetty.io/pricingread 2026-09-21

The Jetty Runbook Directory lets a customer pick a published runbook and run it from their own account with nothing to install, and runbooks can be written, published and shared. Each runbook is a whole workload that sets its own job, what counts as done and its checks.

SourceJetty, jetty.io/docs/get-started and pricingread 2026-09-21

Workloads run on a schedule as well as manually, from the terminal or through the API, and paid plans add webhooks, so work reaches the agent without a person starting it. Jetty supports scheduled workloads and offers a guide to running a runbook on a cron cadence.

SourceJetty, jetty.io/pricing and homepageread 2026-09-21

Customers pick the agent and model pair for each workload, and Jetty says it works with any model. Every paid plan lets customers bring their own provider keys, billed by the provider, and Squadron Platform adds the option to bring your own AI gateway, so the customer controls model choice.

SourceJetty, jetty.io/pricing and homepageread 2026-09-21

Apps connect to Jetty through an API, a TypeScript Client SDK and an MCP server, with API keys on every plan. The tasks API deploys, runs and monitors workloads, and a chat completions layer is compatible with OpenAI.

SourceJetty, jetty.io/docs/get-started, pricing and llms.txtread 2026-09-21

Runbooks carry code checks, checklists and LLM judges that run in flight, and each run is scored against those checks with a verdict and a workload report. The customer evaluates its own agent's work with scored results across many runs. Judges and rubrics come with Squadron Platform.

SourceJetty, jetty.io homepage, docs/get-started and pricingread 2026-09-21

Jetty does not say whether a workload can operate a browser. Playwright and Firecrawl MCP servers are listed as ready on every run of the opencode runtime, but Jetty does not say whether that means a browser running inside its sandboxes.

SourceJetty, jetty.io/llms.txtread 2026-09-21