Skip to main content

​
Bring Your Own Models

On paid tiers, the Sigilix CLI and Deep-Research Chat are not locked to one provider. Point Sigilix at OpenAI’s Codex CLI, Anthropic’s Claude Code, or your own SDK and provider keys, and you pay that provider’s usage directly. The model you bring still fetches Sigilix’s verified earned-context layer, so it inherits the precision of a Sigilix review without re-deriving your repository. This is the whole bet behind bring-your-own-models: the value isn’t the model, and it isn’t the chat loop. The value is the verified context your model fetches. A frontier model with a generic prompt re-discovers your codebase every session and guesses at the edges. The same model pointed at Sigilix’s earned context starts from a repo that has already been indexed, graphed, reviewed, and remembered.
BYO models apply to the CLI and chat surfaces. The hosted PR ensemble runs per-role tuned models with cross-provider fallback so one provider’s outage cannot silence a specialist — see The Ensemble. BYO is a paid-tier feature; availability is per-account during the private beta. Join the private beta or contact support@sigilix.ai.

​
What you can bring

Codex CLI

Use OpenAI’s Codex CLI as the model behind your terminal workflow. Sigilix drives it against the earned-context layer; you authenticate to OpenAI and pay OpenAI directly.

Claude Code

Use Anthropic’s Claude Code as the agent driving the work. Sigilix supplies the verified context; Anthropic bills you for usage on your own account.

Your own SDK / keys

Bring your own SDK or provider keys and route through them directly — useful when you have an existing provider contract, a self-hosted endpoint, or a model not covered above.

​
Why bring your own model

Pay the provider directly

Usage runs on your account with the provider you choose. No model markup from Sigilix on the CLI and chat surfaces — you control the spend.

Use the model you trust

If your team has standardized on a coding agent, keep using it. Sigilix supplies context and gates; the model you bring does the reasoning.

Inherit the precision

Whatever model you bring, it fetches the index, code graph, trust ledger, review memory, and evidence manifests — so it reasons about the whole repo, not a window.

Stay within your boundary

Routing through your own keys means your code goes to the provider you’ve already vetted and contracted with, under their data-use terms.

​
How it fits together

The model is the engine; the earned-context layer is the fuel. BYO swaps the engine without touching the fuel.
1

Hosted reviews build the context

Every PR review deposits a verified layer — index, code graph, trust ledger, review memory, evidence manifests. This is the substrate your model will fetch.
2

You configure your model

Point Sigilix at Codex CLI, Claude Code, or your own SDK/keys, and authenticate to that provider. Sigilix records which model to drive for your CLI and chat sessions.
3

Your model fetches earned context

When you run a CLI review or a chat query, Sigilix expands the request with the earned-context layer before handing it to your model — so it judges and answers against the whole repo.
4

You pay the provider, Sigilix supplies the rigor

Inference billing lands on your provider account. The believability gates, context expansion, and proof-tier receipts come from Sigilix.

​
Configuring a model

The CLI exposes model selection through sigilix config. The exact provider setup depends on which model you bring — typically you authenticate to the provider once, then select it as your active model. 
# Inspect the current model and available providers
sigilix config model

# Select a provider for your CLI / chat sessions (illustrative)
sigilix config model codex
sigilix config model claude-code
sigilix config model custom

# Bring your own provider key (illustrative — store via your env/secret manager)
sigilix config key set
Commands above are illustrative — the precise flags and provider names are surfaced by sigilix config model for your account during the beta. Authentication follows each provider’s own flow (Codex CLI and Claude Code authenticate to OpenAI and Anthropic respectively); for “your own SDK / keys,” you supply the provider credentials.

​
Frequently asked

No. The hosted PR ensemble runs per-role tuned models — a reasoning-heavy model for Metis, a faster high-volume model for Argus, and so on — behind cross-provider fallback so one outage can’t silence a specialist. BYO applies only to the CLI and Deep-Research Chat surfaces. See The Ensemble.
You do — directly to the provider whose model you bring. That’s the point of BYO: usage runs on your own account with the provider you’ve chosen, with no model markup from Sigilix on these surfaces.
Yes. The precision doesn’t come from a particular model — it comes from the earned-context layer and the believability gates. Whatever model you bring fetches that verified context, so it inherits the grounding without re-deriving the repo.
On the CLI and chat, Sigilix uses a default model so you can get going without configuring a provider. BYO is for teams that want to run on their own provider account or standardize on a specific coding agent.
BYO is a paid-tier feature and availability is per-account during the private beta. Join the private beta or email support@sigilix.ai if you’re already in and want it enabled.

Sigilix CLI

Review parity in the terminal — what your model runs against.

Deep-Research Chat

Grounded Q&A over your repo and review history.

Earned Context

The verified layer your model fetches — and why it’s the moat.

Command reference

The CLI commands for review, chat, config, and auth.