Not a pricing page — a promise about costs
WeftKit is a desktop app you build from source today; there are no tiers or license fees to compare here. What this page explains is the one cost the app does meter: what your chosen AI model charges when a repair runs — and how WeftKit makes that impossible to lose track of.
Deterministic runs cost nothing
A replay that passes deterministically involves no model, no tokens, and no spend — it still gets a ledger row, at exactly $0.00. AI can only enter the picture after a step deterministically fails, and only if you enabled repair for the run.
- No AI, no cost — determinism is the default path
- Repair is opt-in per run, with a hard USD limit you set
- If the budget would be exceeded, the request is blocked before it is sent
A run’s ledger entries
Illustrative — amounts are stored as exact 9-decimal USD strings.
How a request gets its price
Three layers, checked in order. The winning price card is snapshotted at run start and stamped into every transaction.
Your overrides
Set a custom price for any model — including one the registry has never heard of. Overrides always win, and setting one registers the model as selectable.
Bundled defaults
WeftKit ships a per-model price registry with published rates for well-known models, each with the provider’s pricing page as its source.
The fallback card
Anything unknown — a local Ollama model, a CLI agent’s internal model, a brand-new release — is priced at $3 per 1M input tokens and $15 per 1M output tokens ($0.30 cached). Never free, never unaccounted.
A durable USD ledger, not a monthly surprise
Every AI request becomes a permanent, append-only record with its token usage, its resolved price card, and where that price came from — bundled default, your override, or the fallback rate.
- Recorded before transmission — the ledger entry is written and flushed to disk before the request leaves your machine
- USD is the immutable base currency; amounts are exact decimal strings, not floats
- The resolved price card is snapshotted into every entry — history is never repriced when rates change
- Deterministic replays appear as explicit $0.00 entries, so the ledger tells the whole story
- When a CLI agent reports its own cost, it is recorded alongside — informational, never a substitute for the metered amount
Budgets that block, not warn
Two independent caps protect you: a per-run hold reserved atomically before the run starts, and a monthly ceiling across everything. Before any request is sent, WeftKit computes a worst-case cost from the price card; if it doesn’t fit, the request never leaves your machine.
Spend nothing by default. Know everything you spend.
Deterministic replays are free; every AI request is priced, capped, and permanently on the record.