Bring your own model (BYOM)

Bring your own model (BYOM) is the seam that lets you connect your own large language model and use it to power every piece of content TrailGuide generates. You supply the provider and the key, and TrailGuide routes every draft, subject line, and variant through the model you chose, not one picked for you.

This matters because content is where a marketing platform touches your brand most directly. If a system is going to write in your voice at scale, you should be the one who decides which model does the writing, on which account, and under which data terms. BYOM makes that decision yours instead of the vendor default.

Any provider, one seam

TrailGuide is provider-agnostic. You can connect a model from any of seven providers, including OpenAI, Anthropic, Google, and Azure, plus gateways to Bedrock and Vertex for teams standardized on AWS or Google Cloud. Whichever you pick plugs into the same internal seam, so the rest of the platform never needs to know who your vendor is.

That single seam is the whole point. Segments, journeys, broadcasts, and one-off sends all call the same generation layer, so switching providers is a settings change rather than a migration. You are never locked into one vendor, because the platform was built to treat the model as a swappable part from the start.

Connect our own OpenAI model as the content generator, using the key in our vault. Then run a test generation to confirm it is wired up correctly.

Your keys, encrypted at rest

When you add a provider key, it is encrypted at rest with AES-256-GCM and decrypted only at the moment a generation runs. Keys are never returned to the browser or to an MCP client in the clear, so the credential you paste in once does not leak back out through the same tools you use to manage everything else.

For teams with stricter obligations, an optional zero-retention flag suppresses content logging entirely. With it on, prompts and completions are not stored, so the text your model reads and writes does not linger in TrailGuide beyond the send itself. You decide whether generation leaves a trail or none at all.

See exactly what your model powers

BYOM ships with a transparency panel so the model is never a black box. It lists every feature the connected model powers, from journey copy to broadcast drafts to variant generation, in one place. Next to it sits a plain-language data-handling statement that describes what is sent to the model, what is stored, and what is not.

The panel exists so that anyone on your team, or an auditor asking how customer content is generated, can get a straight answer from one screen. Transparency about which model runs and what it sees is part of using your own model responsibly.

A test model button lets you fire a single generation on demand and read the response, so you can confirm a newly connected model works before it writes anything a customer will see. When you rotate a key or move to a different provider, you verify in seconds instead of waiting for the next campaign to surface a problem.

Fails closed, never locked in

If a model is misconfigured, the seam fails closed. A bad key, a revoked credential, or a provider outage stops generation rather than silently falling back to some default account or shipping empty copy. You get a clear failure you can fix, not a quiet substitution you discover days later in your metrics.

The same design that fails closed is what keeps you portable. Because the model sits behind a seam and every feature calls it the same way, moving from one provider to another, or bringing a fine-tuned model of your own, changes one connection and nothing else. The model decides what to write; you decide which model does it, and you can change your mind whenever you want.

That portability is why BYOM fits the larger idea that TrailGuide is a marketing harness layered on the stack you already run. Your model is your infrastructure. The harness puts it to work across every send; it does not try to replace it.