AI has become a runaway, unpredictable line item. Outlay attributes every dollar of LLM and coding-agent spend to the work you already plan — tickets, epics, roadmap — forecasts cost by scope, and lets you enforce a budget per program, reallocating compute to the work that matters most. Finally, the fastest-growing line item is as governable as the work that drives it.
“What is this body of planned work costing us in AI, and are we on track?” Today's tools meter infrastructure — API keys, models, dashboards. Outlay meters the work.
See exactly what each ticket, epic, and sprint costs in AI — and which work is about to blow its estimate — the way you already plan effort.
Allocate AI spend to teams and cost centers, set budgets by scope, and get alerts before overspend — not a surprise invoice at month-end.
Connect the tracker and AI tools you already use. Read-only to start — no app rewrite, no prompts leaving your environment.
Every dollar of agent & LLM spend mapped to the ticket, epic, and roadmap it belongs to.
# tag / branch / PR → ticket
fix/PROJ-123 → PROJ-123 → Q3-stability
Predict a quarter's cost from its open scope, the way you estimate engineering effort.
# open roadmap × per-class cost
Q3 forecast: $48k ± $9k
Pace-based alerts flag a team trending over before it blows through — not after.
# projected, not a hard cap
epic Q3-stability on pace $22k · over
Every gateway, observability, and FinOps tool stops at an infrastructure tag. Outlay resolves the git context of each AI call back to the ticket — and budgets it like real work.
Illustrative. Flagged OVER early — driven by one outlier ticket at 11× its class median.
Agentic coding re-sends a large cached prefix every turn, so cache-read tokens — billed at roughly a tenth of base input — dominate the count. Price every token at the base rate, the way most spend tools and spreadsheets do, and the number balloons. We costed our own Claude Code usage both ways.
Measured on our own build sessions:
98% of input-side tokens were cache reads, so the naive method overstates by 7.4×.
Reproduce it yourself: python -m outlay.dogfood --proof-only.
Roadmap and sprint planning shouldn't guess at compute. Hand Outlay the planned work — epics and tickets with their requirements and design docs — and it prices each one against the cost model it learned from your own delivered work, with a confidence range you can budget against.
Illustrative. Each item priced from your own per-work-type history; thin-scope items get a wider band, not a false-precise number.
Group the work that shares a goal — several teams, projects, or work types — into a program, and give it a budget. Outlay alerts before it's blown. And with the opt-in gateway in front of your calls, that budget becomes a true hard cap: an over-budget program is automatically blocked or routed down to a cheaper model — so compute flows to the programs you care about most. As priorities shift, move the budget; enforcement follows.
Illustrative. Platform is over its cap, so new Platform calls route down automatically — freeing budget for the launch. Raise a program's cap to send compute back to it.
LLM and coding-agent spend attributed to the ticket, epic, sprint, and team it belongs to — the allocation finance has never had for AI.
Predict a quarter's AI cost from its open scope, then watch budget-vs-actual burn down in real time.
Pace-based alerts (ok → warn → over) by default, read-only. Or set a program hard cap the opt-in gateway enforces — block or route-down over-budget work automatically.
Every forecast is back-tested on your own closed tickets, leave-one-out — we show the measured error and the sample size, by work type. No vendor benchmark to take on faith.
Jira, Linear, GitHub Issues; Claude Code, Cursor, the API. Reliable attribution even for remote / CI agents via explicit task-tagging.
Attribution runs on metadata — connected with read-only tokens. Prompt content, model outputs, and your API key never leave your environment.
Outlay is a hub, not a silo. Export FOCUS-aligned cost rows for any FinOps/BI tool, pull them programmatically through a read-only API, and stream your audit log to your SIEM — all reconciled to the provider invoice, to the cent. No lock-in, ever.
Per-ticket charge rows using the FinOps Open Cost & Usage Spec column names — load straight into any FOCUS-aware FinOps or BI tool.
# spec column names · per-ticket rows
BilledCost,ServiceCategory,Tags{team,work_type}
A token-authed, rate-limited endpoint returns your latest report as JSON for any warehouse or BI pipeline — the same attributed numbers as the console.
GET /api/v1/spend
→ { total_usd, rows[], data_quality }
Stream every security event to Splunk, Datadog, or S3 — poll the audit API with a cursor for gap-free ingestion, or export CSV for GRC.
GET /api/v1/audit?since=cursor
Plus HMAC-signed webhooks for budget & anomaly events (retried, with a delivery log), and a printable month-end close pack — emailed automatically with the FOCUS CSV attached.
Outlay reads spend from every major provider and connects it to the systems where the work actually happens — read-only, no app rewrite, no prompts leaving your environment.
Reconciles against your provider invoice — Anthropic, AWS Cost Explorer, GCP Cloud Billing, OpenAI Costs — so finance trusts the number, not a token estimate. Connect your stack →
Most spend tools see everything you send. Outlay is built so the sensitive data physically can't reach us — purpose-built for teams that can't let prompts leave their environment: healthcare, legal, and financial services.
FinOps suites allocate cloud bills. Observability tools trace API calls. Native consoles cap per seat. None attribute AI spend to the work you plan, or budget it by scope — because they live at the infrastructure layer and can't see your roadmap. And the ones that compute from tokens price cache reads at full rate, so their per-team numbers can run several times too high.
| Outlay | FinOps suites | LLM observability | Native consoles | |
|---|---|---|---|---|
| Attributes spend to tickets, epics & roadmap | ✓ | ✗cost-center tags only | ✗spans / agent-runs | ✗seat / team |
| Forecasts cost from planned scope | ✓ | ~cloud trend forecasts | ✗ | ✗ |
| Pace-based budget guardrails by scope | ✓ | ~account budgets/alerts | ✗ | ~per-seat caps |
| Cache-aware costing (per token class) | ✓cache reads at ~0.1× | ~reads the bill, can't attribute | ✗tokens at a flat rate | ~provider total only |
| Reconciled to the provider invoice | ✓Anthropic, Bedrock, Vertex, OpenAI | ✓it is the bill | ✗ | ✓single provider |
| Forecast accuracy back-tested on your own work | ✓leave-one-out, with sample size | ✗ | ✗ | ✗ |
| Prompts never leave your environment | ✓ | ✓reads billing, not content | ✗traces prompts/outputs | ✓ |
Where they fit: keep a FinOps suite for cloud, and observability for debugging quality. Outlay is the layer they can't be — AI spend attributed to the work you plan, forecast by scope, and held to budget. See the full comparison →
Outlay is live and onboarding its first customers now. Your first weeks are free — and because you're early, you get a team that ships fast: tell us what you need and we build it.
A focused start: connect one tracker + your AI tools, get your real spend mapped to tickets and a budget forecast in weeks — read-only, no app rewrite.
Early-customer pricing for ongoing budgeting & governance — talk to us.
It resolves the work context of each AI call from the most reliable signal available — an explicit task tag from your agent launcher or CI, the git branch, the PR's closing-issue link, or a commit trailer — and maps it to the ticket, epic, and roadmap in Jira/Linear/GitHub. Where a team already links work to tickets it's automatic; where they don't, a one-line explicit tag makes it reliable — even for remote/CI agents where the branch is detached.
No. Outlay connects with read-only tokens and reads metadata — task category, token counts, ticket IDs, and your provider's usage data. Prompt text, model outputs, and your API key never reach our servers.
Both — your choice, per program. By default they're pace-based projections: a scope on track to exceed its budget is flagged warn (and over once it crosses), so a lead acts while there's time — Outlay never touches your traffic. For a true hard cap, set a program to enforce: with the opt-in gateway in front of your calls, an over-budget program is automatically blocked or routed down to a cheaper model. That gateway is the one mode where we're in your path, it's entirely opt-in, and it fails open — a control-plane blip never blocks your traffic, only a budget you set being exceeded.
We don't ask you to trust a vendor benchmark. Outlay back-tests its forecast on your own closed tickets, leave-one-out — hide a ticket, predict it from the rest, compare to what it actually cost — and shows the measured error and the sample size, by work type. As more work closes, the number sharpens.
Yes — that's the forward estimator. Hand it a planned backlog (epics/tickets with their requirements, design docs, and story points) and it prices each item against the cost-per-work-type model learned from your own delivered work, returning a per-item and total estimate with a low–high confidence band. Under-specified items are flagged to tighten rather than guessed, so you can budget an epic or a sprint before committing to it.
Trackers: Jira, Linear, GitHub Issues. AI tools: Claude Code, Cursor, and the Anthropic API (per-call usage + the org admin API for reconciliation). Read-only to start — no app rewrite.
Outlay is live and onboarding its first customers; your first weeks are free and we're setting early-customer pricing now. Talk to us to get started.
See what your AI compute costs per ticket, epic, and team — and where it's about to go over. We're onboarding our first customers now.
Read-only to start · prompts never leave your environment · no app rewrite