Questions engineers ask before they ship.

Trellis is the runtime that sits between your AI agents and your systems: it keeps runs durable, enforces your rules, tests changes before they ship, and records a priced trace of every step.

If someone could write the steps down in advance, build a pipeline. It will be cheaper, faster, and easier to debug. Agents earn their cost only when the next step depends on what the last one found. Our essay on when not to use an agent walks through the test.

About ten minutes with the quickstart. You install the SDK, define a small agent, run it in the dev sandbox, and open its trace. The sandbox uses mock tools, so nothing real is touched.

There are TypeScript and Python SDKs, a CLI, and an HTTP API underneath them. If you already use an open-source agent framework, you can keep it and send its runs through Trellis.

Any provider with an API. Bring your own keys, or use our metered pass-through at provider list price. You can switch providers per agent, per step, or per eval candidate.

Two meters and a pass-through. You pay a small orchestration fee per run and $0.08 per 1,000 guardrail checks. Deeper trace retention is a flat add-on per run. Model tokens pass through at provider list price with no markup.

Yes. The dev tier includes 1,000 runs a month, sampled traces, every guardrail mode, and the CI gate. It has no time limit and needs no card.

No. Seats are free and unlimited on every plan. We want more of your engineers reading traces, not fewer.

The run stops before the step that would cross the cap and takes the fallback branch you chose: escalate, park, or fail. Workspace caps stop new runs and page whoever is on call. You never find out from an invoice.

No. Trellis Agents is a fictional company and this is a sample site. Every price, rate, and metric is an illustrative sample, and model prices are not quotes from any real provider.

You choose a US or EU region when you create a workspace. Traces, eval datasets, and audit logs stay in that region.

No. Your traces, prompts, and eval data are used to run your workspace and nothing else. We do not train or fine-tune any model on customer data.

Yes, per workspace: 14 days sampled, 30 days full, or 90 days forensic with full payloads. Forensic retention supports customer-managed encryption keys, and every retention change is audit-logged.

Yes. We fill in your questionnaire, share a data-flow diagram, and walk your team through access, retention, and key controls on a call. This sample site does not claim any audit or certification.

Enterprise workspaces can run in a dedicated single-tenant region. A fully self-hosted version is not offered today, and we would rather say so than promise a date.

In-flight runs stop at their last checkpoint and resume when service returns, with no repeated side effects. You can also set the SDK to fail open for read-only agents that must keep answering.

Start it in observe mode, read the would-have-blocked report for a week, move to dry-run so reviewers rehearse, then switch to enforce. Each change is one line and lands in the audit trail.

Run trellis eval --gate in CI. If a candidate prompt, tool, or model drops below your thresholds, the command exits with code 1 and the pipeline stops, just like a failing unit test.

Yes. Every span exports over OpenTelemetry to your collector or to common dashboards. Trellis is meant to be one source of truth, not the only place your data can live.

We follow semantic versioning. Breaking changes come with a migration guide and at least 90 days' notice before the old version turns off, and every release is listed in the public changelog.