/synthex:configure-multi-model/synthex:configure-multi-model
SetupWizard that detects installed model CLIs, checks auth, and enables multi-model review.
Invocation
Run from inside Claude Code in any project where Synthex is installed:
/synthex:configure-multi-modelSome commands accept parameters; see the source-of-truth file linked below for the full parameter list and workflow.
Lifecycle phase
Setup. One-time project scaffolding.
Related agents
No agents declare this command's lifecycle phase. The command runs structurally — delegating to whichever agents the command's workflow names directly.
Source
The command's parameters, workflow, and sub-agent invocations are defined in markdown at bluminal/lumenai/blob/main/plugins/synthex/commands/configure-multi-model.md.
Last modified:
What it does, end-to-end
/synthex:configure-multi-model/synthex:configure-multi-model is the setup wizard for the multi_model_review block
in .synthex/config.yaml. You can re-run it any time. /synthex:init/synthex:init runs the same
step on a new project. Each run:
- Checks what's already there. If multi-model review is already on, it shows your current
reviewers, both per-command switches, and the aggregator, and asks what to do (see Re-running it).
Otherwise it starts fresh. Other keys in an existing
multi_model_reviewblock are kept. - Detects installed CLIs. For each supported CLI it checks, in parallel, both that the binary is installed and that it's logged in. Only the exit code counts.
- Asks you to pick one of three options, listing what it found.
- Shows the data-transmission warning before it writes
enabled: true. - Writes the config and runs preflight, then prints a summary such as
3 reviewers configured, 3 available, 3 families, aggregator: codex-review-prompter. It also createsdocs/reviews/for audit records.
What it detects
Each login check is the same one the matching adapter runs, so the wizard never offers a CLI that the adapter would then reject, or the reverse.
| CLI | Adapter (reviewers entry) | Installed check | Login check |
|---|---|---|---|
codex | codex-review-prompter | which codex | codex login status, run with CODEX_API_KEY and OPENAI_API_KEY unset |
gemini | gemini-review-prompter | which gemini | GEMINI_API_KEY or GOOGLE_API_KEY set; otherwise gcloud auth list |
ollama | ollama-review-prompter | which ollama | curl -sf http://localhost:11434/api/tags (server running) |
llm | llm-review-prompter | which llm | None. Keys are set per provider plugin, so an installed llm counts as logged in |
aws | bedrock-review-prompter | which aws | aws sts get-caller-identity --output text |
claude | claude-review-prompter | which claude | claude auth status |
Results fall into three groups: installed and logged in, installed but not logged in, and not
installed. CLIs that are installed but not logged in are listed separately with a fix, for
example gemini — set GEMINI_API_KEY or run gcloud auth login to enable.
The three options
- Enable with detected CLIs. Writes
multi_model_review.enabled: true, areviewerslist with the adapter names of only the CLIs that passed both checks, and turns onper_command.review_code.enabledandper_command.write_implementation_plan.enabled. If none passed, the option is still offered (labelled "none currently authenticated") so you can fix logins afterwards. - Enable later (show snippet). Prints a commented-out
multi_model_review:block with your detected CLIs and both per-command switches, for you to paste in when ready. Nothing is written. - Skip. Writes nothing. Multi-model review stays off.
If preflight fails, the wizard prints how to fix it but doesn't stop. During /synthex:init/synthex:init, the rest of setup carries on.
The data-transmission warning
Before it turns the feature on, the wizard shows:
Multi-model review sends your code, diffs, and (for write-implementation-plan) draft plan markdown to the configured external CLIs. Each CLI relays this content to its underlying provider (OpenAI, Google, etc.) per the provider's terms of service. Synthex does not store, log, or modify this content beyond what's needed to invoke the CLI.
If you need to keep all review content local, configure ONLY local-model adapters (Ollama, Bedrock with on-prem model) and remove hosted-model adapters from
reviewers.
Parameters
| Parameter | Description | Default | Required |
|---|---|---|---|
config_path | Where the config file lives | .synthex/config.yaml | No |
Re-running it
When multi-model review is already on, the wizard shows your current settings. If both
per-command switches are false, it also notes that no command uses multi-model review yet,
and that re-running the wizard turns it on. Then it offers:
- Re-run the wizard. Detect CLIs again and replace
multi_model_review.reviewersand the two per-command switches. The data warning is shown again before anything is written. - Reset to disabled. Sets
multi_model_review.enabled: falseand keeps the rest of the block, including the per-command switches. The master switch overrides them. The explicitfalserecords that you opted out. - Leave as-is. Exits with no changes.
Per-command switches
/synthex:review-code/synthex:review-code and /synthex:write-implementation-plan/synthex:write-implementation-plan use
multi-model review only when the master switch and their own per-command switch are on. Both
per-command switches default to false. The Enable with detected CLIs option sets both to
true for you. Before Synthex 2.5.3 it wrote only the master switch, so multi-model review never
actually ran after the wizard. If you ran the wizard on an older version, re-run it or add:
multi_model_review:
per_command:
review_code:
enabled: true
write_implementation_plan:
enabled: trueTo keep one command native-only, set its switch back to false. Or leave both off and pass
--multi-model for a single run. See
Multi-model review for how each command
decides.
What it does not do
- Ask for or store API keys. Each CLI handles its own login (
codex login,gcloud auth login,ollama serve, and so on). Config files never hold keys. - Set models, permission modes, or other per-reviewer settings. Edit those keys yourself. Put
model choices (
multi_model_review.per_reviewer) in your user config at~/.config/synthex/config.yaml. - Delete the block when you disable it. Reset keeps the block and sets
enabled: false.
See also
- Multi-model review — how it works, adapters, and failure behavior.
- Configuration — why
enabled: falsein any layer wins. - /synthex:review-code/synthex:review-code — the main command that uses it.