VAMOS Assist¶
VAMOS Assist provides a no-code workflow for planning, materializing, and running experiments from the command line. You can use deterministic template-first planning or optional LLM-assisted planning. Every step writes reproducible artifacts to disk, and generated configs are validated against the ExperimentSpec v1 schema before execution.
Quick Start (Template-First, No API Keys)¶
vamos assist go "template-first example" --template demo --smoke
This path does not require OpenAI, external API keys, or network calls.
Example (Template-first)¶
vamos assist go "template-first example" --template demo --smoke --json
Expected JSON output (example):
{
"plan_dir": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx",
"plan_paths": {
"catalog": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\catalog.json",
"config": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\config.json",
"plan": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\plan.json",
"prompt": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\prompt.txt"
},
"project_dir": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\project",
"recommended_commands": [
"vamos --config results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\project\\config.json",
"vamos --config results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\runs\\run_YYYYMMDD_HHMMSS_xxxxxx_smoke\\execution_config.json"
],
"run": {
"execution_mode": "in_process",
"exit_code": 0,
"execution_config_path": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\runs\\run_YYYYMMDD_HHMMSS_xxxxxx_smoke\\execution_config.json",
"run_dir": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\runs\\run_YYYYMMDD_HHMMSS_xxxxxx_smoke",
"run_report_path": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\runs\\run_YYYYMMDD_HHMMSS_xxxxxx_smoke\\run_report.json",
"status": "ok"
}
}
Note: execution_mode can be "in_process" or "subprocess" (fallback). The actual mode is recorded in run_report.json.
Auto Planning with OpenAI (Optional)¶
Install optional dependencies:
pip install vamos-optimization[openai]
Alternative:
pip install openai
Set your API key:
Windows:
setx OPENAI_API_KEY "..."
macOS/Linux:
export OPENAI_API_KEY="..."
Run diagnostics:
vamos assist doctor
Run auto planning:
vamos assist go "I have 10 continuous variables, minimize 2 objectives..." --mode auto --provider openai --smoke --json
VAMOS requests structured output from the provider (strict schema) and validates the resulting config before running.
Example (Auto planning with mock provider — no API keys)¶
vamos assist go "optimize a 2-objective continuous benchmark quickly" --mode auto --provider mock --smoke --json
Expected JSON output (example):
{
"plan_dir": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx",
"plan_paths": {
"catalog": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\catalog.json",
"config": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\config.json",
"plan": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\plan.json",
"prompt": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\prompt.txt"
},
"project_dir": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\project",
"recommended_commands": [
"vamos --config results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\project\\config.json",
"vamos --config results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\runs\\run_YYYYMMDD_HHMMSS_xxxxxx_smoke\\execution_config.json"
],
"run": {
"execution_mode": "in_process",
"exit_code": 0,
"execution_config_path": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\runs\\run_YYYYMMDD_HHMMSS_xxxxxx_smoke\\execution_config.json",
"run_dir": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\runs\\run_YYYYMMDD_HHMMSS_xxxxxx_smoke",
"run_report_path": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\runs\\run_YYYYMMDD_HHMMSS_xxxxxx_smoke\\run_report.json",
"status": "ok"
}
}
Example (Auto planning with OpenAI provider)¶
Install optional dependency support:
pip install vamos-optimization[openai]
Alternative:
pip install openai
Set your key:
Windows:
setx OPENAI_API_KEY "..."
macOS/Linux:
export OPENAI_API_KEY="..."
Run diagnostics:
vamos assist doctor --json
Example diagnostics snippet:
{
"openai": {
"sdk_available": true,
"api_key_set": true,
"model": "gpt-5.2",
"temperature": 0.2,
"max_output_tokens": 900
}
}
Run auto planning:
vamos assist go "I have 10 continuous variables, minimize 2 objectives..." --mode auto --provider openai --smoke --json
Expected JSON output (example):
{
"plan_dir": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx",
"plan_paths": {
"catalog": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\catalog.json",
"config": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\config.json",
"plan": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\plan.json",
"prompt": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\prompt.txt"
},
"project_dir": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\project",
"recommended_commands": [
"vamos --config results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\project\\config.json",
"vamos --config results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\runs\\run_YYYYMMDD_HHMMSS_xxxxxx_smoke\\execution_config.json"
],
"run": {
"execution_mode": "in_process",
"exit_code": 0,
"execution_config_path": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\runs\\run_YYYYMMDD_HHMMSS_xxxxxx_smoke\\execution_config.json",
"run_dir": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\runs\\run_YYYYMMDD_HHMMSS_xxxxxx_smoke",
"run_report_path": "results\\assist_plans\\plan_YYYYMMDD_HHMMSS_xxxxxx\\runs\\run_YYYYMMDD_HHMMSS_xxxxxx_smoke\\run_report.json",
"status": "ok"
}
}
Note: OpenAI usage is billed to the key owner (BYOK). See Billing (BYOK). In auto mode, VAMOS sends your prompt plus compact planning context (templates/algorithms/kernels/allowed override keys), not datasets or arbitrary local files.
Billing (BYOK)¶
When using --provider openai, OpenAI API usage is billed to the account/organization associated with the OPENAI_API_KEY on the machine running VAMOS.
VAMOS does not ship any API key.
Privacy¶
In auto mode, VAMOS sends the user prompt and compact planning context (for example templates, algorithms, kernels, and allowed override keys, potentially truncated for size) to the provider. VAMOS does not automatically upload datasets or local files.
Plan trace files (provider_request.json, provider_response.json) are written to the plan directory and do not store secrets such as OPENAI_API_KEY.
Commands¶
vamos assist catalog: show available algorithms, kernels, operators, and templates.vamos assist plan: create a plan directory (template mode or auto mode with provider).vamos assist apply: turn a plan into a runnable project directory.vamos assist run: execute a plan/project config; supports--smokeand--smoke-evals.vamos assist explain: summarize plan/run defaults, paths, and overrides.vamos assist doctor: show environment, catalog, and provider readiness diagnostics.vamos assist go: high-level happy path (plan -> apply -> optional smoke run).
Artifacts (what gets created)¶
Plan directory (assist plan / assist go):
prompt.txtcatalog.jsonconfig.jsonplan.jsonprovider_request.jsonandprovider_response.json(auto mode only)
Project directory (assist apply / assist go):
project/config.jsonproject/README_run.md
Run directory (assist run / assist go --smoke):
runs/run_*/execution_config.jsonruns/run_*/run_report.json- results under the run output root (
defaults.output_root, typically inside the run directory)
run_report.json records execution_mode (in_process or subprocess fallback).
Troubleshooting¶
- Missing OpenAI SDK: install optional deps with
pip install vamos-optimization[openai](orpip install openai). - Missing API key: set
OPENAI_API_KEYand open a new shell session if needed. - Run diagnostics first:
vamos assist doctor. - Invalid auto overrides: VAMOS filters/rejects unsupported keys and records requested/applied/rejected overrides in
plan.json.
For development installs, you can also use the same vamos assist ... commands after installing the repo in editable mode.