saascode
saascode

CallThrift

A multi-provider inference FinOps gateway that attributes every model call to customer, agent, task, and tool; enforces versioned budgets; compares outcome-normalized routes; and recommends or executes cheaper paths only under explicit quality and reliability policy.

Genesis score7.05/10
Make CallThrift real.0/500
500 more votes and CallThrift is authorized for build.
0%500 to authorize
Backing is the vote. When an idea crosses 500, we pull it into the build pipeline and ship it for real — the votes decide what gets built next, not an editor.
The opportunity
10M calls/moClosest free tier
90 daysObserved retention
5 fieldsCost ledger key
The case

An agent builder can receive one aggregate provider invoice while individual customers, agents, tools, retries, and background tasks consume wildly different amounts. A cheap-route recommendation is dangerous if it ignores output quality, latency, data policy, or the cost of a failed task. CallThrift creates a five-part cost record around each call, reconciles it with provider billing, and ties routing decisions to outcome and policy. A budget kill switch limits future spend; it cannot reverse a call already accepted or prove savings unless the baseline, quality, and avoided cost are measured.

Who pays — and why

The engineering, AI-platform, or finance owner at an agent company with several providers and customers, meaningful inference spend, and no reliable allocation from a model call through a tool invocation to the customer outcome it served.

What it unlocks
A per-customer cost-and-outcome corpus that can distinguish genuinely cheaper task routes from routes that merely have a lower token price
A request-linked ledger across customer, agent, task, tool, provider call, retry, and billing record that supports chargeback and budget policy
An aligned-savings commercial hypothesis whose baseline and quality controls are independently inspectable rather than hidden inside a zero-markup claim
How Genesis scored it
7.05across seven criteria
tension 6temporal 8blindspot 5buyer 8leverage 8convergence 5why-not 8
8
Temporal window

The 2026 protocol-tooling wave and several new cost-tracking projects show an active infrastructure window.

8
Buyer persona

Agent-company engineering and finance owners have a direct need to allocate and control multi-provider spend by customer and task.

5
Convergence

One cross-reference, four inbound connections, and one direct link form a limited local cluster.

Why it scored well

The agent-builder buyer and cost-allocation job are clear, active open and hosted products validate per-agent and per-tool telemetry, and request-linked attribution plus routing can compound through spend patterns. The product is software-delivered.

What's holding it back

The category is more crowded than the initial synthesis suggested: one hosted competitor already provides per-tool and per-customer attribution, while open tools cover budgets, anomaly detection, and route recommendations. Routing plus verified savings is the remaining slice, and no structural incumbent handicap is proven.

Signals detected4 sources crossed
SignalMCPSpend product research

Signalopenclaw-cost-tracker-mcp and agent-costcenter-mcp repository research

SignalZop.dev architecture research

Signal2026 competitor synthesis

Direction briefcallthrift.md
callthrift.md
Want this pointed at your vertical?Point Genesis at your own market and constraints — it invents adjacent, fork-ready ideas, private to you before they hit the public feed.

Discussion

?

No comments yet — be the first to weigh in.