The advisor

The advisor gives the agent someone to ask. When it reaches a hard decision, such as two approaches that both look reasonable, a bug that keeps resisting, or a design trade-off, the agent can send the question to a second model and get a short, reasoned recommendation back. The agent then decides and acts itself.

How it works

When the advisor is on, the agent gets an extra tool named Advisor. It uses the tool sparingly, on decisions where a second opinion is worth the wait.

1. The agent writes a question and, optionally, the context the advisor needs: code, error messages, facts it has found.
2. Syntic Code sends both to the advisor model with instructions to give a brief, reasoned recommendation and to say what the agent should check first.
3. The answer comes back into the conversation labelled with the model that gave it, for example Advisor (Amara Deep) says: …, so the transcript shows whose advice it was.

The advisor only advises. It sees what the agent sends, nothing more: it cannot run tools, read files or change anything. Because of that, calling it never needs a permission prompt.

Limits

Turn it on

/advisor with no argument (or /advisor status) tells you whether it is on and which model it would ask:

> /advisor
Advisor: off (would ask Syntic Code). /advisor on turns it on; /advisor <model> picks the model and turns it on.

A /advisor choice wins over the environment variable for the rest of the session.

Choose the advisor model

The advisor asks, in order of preference:

1. the model you named with /advisor <model>;
2. SYNTIC_ADVISOR_MODEL;
3. the session’s main model.

/advisor default drops the session choice and goes back to the variable or the main model. Model names are the ones /model lists, such as Amara Deep; names with spaces work as typed.

A useful pattern is to work on a quick model and ask a deeper one for advice:

SYNTIC_ENABLE_ADVISOR=1 SYNTIC_ADVISOR_MODEL="Amara Deep" syntic --model "Syntic Code"

The advisor model must be one your plan includes; see Plans and models.

Turn it off

/advisor off turns it off for the session. While it is off, the Advisor tool is not offered to the agent at all, so it costs nothing. To keep it off by default, leave SYNTIC_ENABLE_ADVISOR unset.

Cost

Each consultation is an extra model request: the question and context go in, the answer comes out. It runs on the advisor’s model and uses your plan like any other request. A handful of consultations in a long session is small next to the main conversation; see Usage and billing.

When it goes wrong

Advisor, subagents or a bigger model?