Cost Plane. See what AI costs.
And what it returns.
Spend attributed across retries, context, and routing · so you see which workflows pay off and which fail. The knobs to drive cost down, and a defensible path to ROI.
The token bill arrives. Nobody can explain it.
AI rollouts grow per-seat costs faster than per-seat outcomes. Finance asks where the money went. Engineering shrugs. The CFO escalates. Nobody has a dashboard.
Retries hide inside the bill
Every wrong answer the user reprompts. Every retry is a new charge. The accuracy problem hides inside the cost problem · and the CFO sees only the bill.
Context bloat compounds
Tool definitions consume tokens before the question is asked. Long histories pad every query. The agent rediscovers schema on every call.
Premium routing by default
Every query goes to the most expensive model · even the simple ones. Without routing logic, the spend tracks the worst-case prompt, not the workload.
See where every token goes.
By user, by team, by query.
Cost Plane attributes the bill across the four dimensions that actually drive it · retries, context, routing, and productive work. Then it shows you, with real numbers, where the waste is.
The dashboard
A representative 30-day view from a mid-size enterprise rollout. Notice that 92% of spend is waste · the bottom 8% is the productive part.
Visibility motivates the fix. We don't ask anyone to change anything until they can see where the spend is going.
Three dials. Real impact.
Once you can see the spend, Cost Plane gives you three controls · each one targeting a specific share of the bill.
Cut retries
Use seeded context and Connections caching so the model gets the right answer the first time. Fewer reprompts. Fewer charges.
Trim context
One CLI tool definition replaces 50+ per-system MCPs. Workspace artifacts (joins, queries) replace ad-hoc context dumps.
Route smarter
Set routing rules. Use cheaper models for simple queries. Use premium models only when the workload demands it.
Set the ceiling. The workspace enforces it.
Per-team, per-user, or per-workflow caps. Soft alerts. Hard cutoffs. Renewable budgets per billing cycle.
Cap a team's monthly spend
Set $5K/month for CS Ops. Get a Slack alert at 80%. Hard-cap at 100%, with optional manager override.
Catch outlier users
Any user burning >5× the team median triggers an investigation alert. Surface the workflow that's doing it.
Bound experimental skills
New skills get a probation budget. Cost-Plane gates production rollout until the per-query cost is below a threshold you set.
Track model spend separately
Separate budgets for Claude, ChatGPT, your in-house model. Switch routing rules without re-architecting your skills.
Start with visibility. Then provide the control knobs.
You can't fix · or let the customer fix · what you can't see. Visibility motivates the customer to reach for the optimization knobs in the first place. We measure and attribute cost before we ask anyone to change anything.
One platform. Three products.
Connections
One CLI fans out to 50+ enterprise systems. The agent loads one tool. The workspace handles the routing.
Connections →Product 02Sandbox
A governed runtime where AI does the work. Scoped credentials, audit, SIEM logs, controls enforced by architecture.
Sandbox →PlatformThe Workspace
The runtime, the context layer, the security perimeter. All three products on one architectural foundation.
Platform overview →Common questions about Cost Plane.
Do you charge for tokens?
No. You bring your own model API keys (Anthropic, OpenAI, etc.) and pay them directly for inference. Cost Plane's job is to make those tokens go further · by showing you exactly where every one of them goes and giving you the controls to cut waste.
How granular is the cost attribution?
By user, by team, by workflow, by query. Split across four dimensions · retries, context length, premium routing, and productive work. The same view drills down to a single query if you need to find the outlier user or the runaway workflow.
Can I set budgets per team or workflow?
Yes · per-team, per-user, per-workflow, and per-model budgets. Soft alerts at any threshold, hard cutoffs at 100%, with optional manager overrides. Renewable monthly or quarterly per billing cycle.
What models does it support?
Every frontier model · Claude, ChatGPT, Gemini, plus open-source models you self-host. Cost attribution works the same way across vendors. Routing rules let you send simple queries to cheaper models and reserve premium models for complex workloads.
How does smart routing work?
You define routing rules · based on prompt complexity, user role, workflow, or budget remaining. Cost Plane evaluates each query against the rules and dispatches to the right model. Override per-query if needed.
Can I see the data in my own BI tool?
Yes. Cost Plane exposes attribution data via API and as structured SIEM events. Stream it to Snowflake, BigQuery, Splunk, Datadog, or pipe it into your own dashboards. The data is yours.
Does it work with my existing AI surfaces?
Yes. Cost Plane tracks token usage across Claude, ChatGPT, Cursor, Copilot, Codex, Slack, and custom apps using the workspace. Same attribution model regardless of where the query originated.
How does it compare to other cost tools like Helicone or Portkey?
Helicone and Portkey are LLM-observability tools · they sit between your code and the model. Cost Plane is part of the MarcoPolo workspace · it sees cost AND the workflow context, which means attribution by user, team, or workflow (not just by model). For pure model observability, Helicone is great; for enterprise cost governance with workflow context, Cost Plane.
Token attribution that lands. Defensible cost math.
Attribute the last 30 days in an hour.
Free Cost Plane attribution on your existing token spend. Bring your model API keys; we'll show you where the money went.
