[NO MATCHING SECTIONS]
Configuration
All settings are in kineti.toml in your project folder — model names,
prices, spend limits, run mode, and sign-in. Data, not code. Providers use the OpenAI format; Gemini
and Grok use one client today.
If the file is missing or has an error, Kineti uses built-in defaults and shows a warning. Settings may be missing, but checks still run.
01COMPLETE SCHEMA
# ── Providers ────────────────────────────────────────────────────────
[providers.<name>] # any name; referenced by --provider
base_url = "https://…" # required — OpenAI-compatible endpoint
api_key_env = "NAME_OF_ENV_VAR" # required
default_model = "model-id" # required
price_per_1m_input = 0.0 # optional, USD per 1M input tokens (default 0.0)
price_per_1m_output = 0.0 # optional, USD per 1M output tokens (default 0.0)
# Optional OAuth2/PKCE per provider — token wins over env key:
[providers.<name>.auth]
client_id = "kineti-cli"
authorize_url = "https://idp.example.com/authorize"
token_url = "https://idp.example.com/token"
scopes = "openid profile" # optional
# ── Execution ────────────────────────────────────────────────────────
[execution]
mode = "single" # "single" | "swarm"
max_parallel_workers = 4 # dependency-wave width in swarm mode
worker_isolation = "auto" # auto | git | scratchpad
# ── Limits ───────────────────────────────────────────────────────────
[limits]
global_usd = 50.0 # whole-run ceiling
per_stage_usd = 10.0 # ENFORCED per-stage ceiling; 0 disables
max_worker_usd = 0 # swarm-only per-worker ceiling; 0 disables
context_char_budget = 24000 # context integrity budget
verify_command = "" # proof the ship gate binds to, e.g. "./greeting.sh"
# ── Clean files ──────────────────────────────────────────────────────
[clean_files]
forbid = [] # extra case-insensitive forbidden substrings (team/client/project names) 02DEFAULTS CHEAT-SHEET
| KEY | DEFAULT | NOTES |
|---|---|---|
| limits.global_usd | 50.0 | stops the run when spend goes over |
| limits.per_stage_usd | 10.0 | limit per step; 0 turns it off |
| limits.max_worker_usd | 0 (off) | limit per worker; spend check still stops overuse |
| limits.context_char_budget | 24000 | characters |
| execution.mode | "single" | |
| execution.max_parallel_workers | 4 | |
| execution.worker_isolation | "auto" | git repo → separate copy, else temp folder |
| providers.gemini | included | GEMINI_API_KEY, free, 0 cost |
| providers.grok | included | XAI_API_KEY, 12.5/25.0 per 1M tokens |
03HOW TO RESUME AFTER STOP
If you go over any spend limit, the run stops. To start again, create
.kineti/spend.reset by hand. There is no flag, API, or setting that
restarts it for you.