For finance and finops
What does each run cost, and who holds the budget?
The questions a review asks of Promenaut, answered from the platform itself, and everything built for them.
The questions, in the order they come.
At the model call. Every call is recorded with its input and output tokens, latency, success or failure, and the organisation and workflow it belongs to.
- Model usage meteringEvery model call that completes is recorded with input and output tokens, latency, and the organisation and workflow it belongs to.
- Per-run cost attributionCost is recorded per model call and as cost legs per stage of work, so spend attributes to the agent and the run that incurred it rather than arriving as one undifferentiated bill.
The full answer is in How Promenaut Accounts for Every Run.
In credits, set by the agents rather than by the tokens they consume.
- A run is charged by its agents' tiers, not its tokensEach completed agent run is charged a fixed number of credits whatever tokens it consumed: one, three, seven or fifteen set by the agent's tier, or a whole number of credits an administrator has set for that agent, which takes the place of the tier.
- Agent credit tiersEvery agent carries a credit tier, so the cost of running a workflow can be reasoned about from its composition before it runs rather than only after.
- Credit ledgerConsumption is recorded against a credit ledger, so an organisation is charged in credits rather than in raw model spend.
The full answer is in How Promenaut Accounts for Every Run.
Yes.
- What this will cost, before you run itBefore a workflow is launched the platform resolves the agents it would run, sums their credit tiers, converts the total to money at the enterprise contracted rate, and reports whether it fits the annual pool and what proportion of the pool it would consume.
- Estimates learn from completed runsEach night the platform recomputes, for each workflow variant with at least five completed runs on its default models, the median and ninetieth-percentile credits, model cost and duration, and the pre-flight estimate uses them in place of the declared figure once they exist.
The full answer is in How Promenaut Accounts for Every Run.
Through business units.
- Budgets allocated to business unitsAn enterprise's annual credit pool is distributed to business units, one for each organisation on the platform, and credits can be reclaimed from a business unit back to the pool.
- Credits move between the pool and business units as paired entriesA budget owner allocates credits from the enterprise pool to a business unit and can reclaim them.
- A warning when a business unit reaches its thresholdA business unit can carry a warning threshold.
The full answer is in How Promenaut Accounts for Every Run.
On the contract's schedule, and by signed file.
- The pool is funded on the contract's scheduleThe annual credit pool is granted from the contract on its anniversary each year and a monthly exploration allowance on the first of each month, each as a ledger grant that can be issued only once for its period.
- Credits provisioned from a signed fileAn administrator applies a signed credits file to a deployment, either to set its contract terms and annual pool or to add a top-up.
The full answer is in How Promenaut Accounts for Every Run.
Finance access is split into distinct permissions for reading, viewing chargeback, managing budgets and managing cost centres, so a finance reader, a budget owner and an auditor are different levels of access.
- Separated finance permissionsFinance access is split into distinct permissions for reading, viewing chargeback, managing budgets and managing cost centres, so a finance reader, a budget owner and an auditor are different levels of access rather than one shared role.
- Least-privilege reporting accessThe credit usage report and the budget tree can be read by a non-human service account holding a scoped, expiring token and a read-only role that grants nothing else, so a reporting integration never needs broad access.
- Administrative audit trailSecurity-relevant administrative changes are written to a hash-chained audit log with the actor, address and session recorded, covering users, roles, teams, access policies, single sign-on, security mode, cost centres, each agent's knowledge envelope and service identity, the definitions of domain, platform and service teams, and workflow configuration settings.
The full answer is in How Promenaut Accounts for Every Run.
A set of finance views built on the metering the platform already captures: chargeback, cost allocation, a cost explorer with a waterfall breakdown, cost mix, model costs and platform usage.
- FinOps reporting surfacesA set of finance-facing views covers chargeback, a cost explorer that breaks spend down by tenant, team, cost centre, agent, workflow, user or model with a waterfall breakdown, cost mix, model costs, platform usage, rightsizing and renewal forecasting.
- Owner chargeback statementFor a closed period a chargeback statement gives the amount to recharge each owner, a tenant or a cost centre, in credits valued at the enterprise credit rate, with the model dollar cost shown beside it and not summed with it.
- Per-enterprise credit currencyEach enterprise prices its credits in a currency held as a commercial-contract term, set through the signed provisioning file and defaulting to US dollars.
The full answer is in How Promenaut Accounts for Every Run.
Wasted spend is separated from spend in four named categories: runs that failed, runs that completed and produced nothing, the share of a run spent on retries, and attempts inside a retry chain that errored, truncated or timed out, with agents ranked by what they waste rather than what they cost.
- Spend that bought nothingWasted spend is separated from spend, in four named categories: runs that failed, runs that completed and produced nothing, the share of a run attributable to its retries, and attempts inside a retry chain that errored, truncated or timed out.
- Is this agent on the model it needsEvery agent is shown against the models it has actually run on, with runs, successes, failures, total and average tokens, average duration and cost, so an agent sitting on a more expensive model than its work requires is visible rather than inferred.
- Model inventoryThe platform keeps a catalogue of the models it can reach, with lifecycle state and health checked against each agent's primary and fallback models, and a change history of manual edits, additions, deprecations and reactivations found by sync, and each agent's published model changes.
The full answer is in How Promenaut Accounts for Every Run.
Adoption is shown as a lifecycle, from invited to using it habitually, so the stage where people stop is visible.
- Where people fall out of adoptionAdoption is shown as a lifecycle rather than a usage count: invited, joined, signed in at least once, ever ran agent work, and using it habitually.
- Which teams took it up, and where seats sit idleAdoption is broken down per team of people, with seats, active seats, dormant seats, teams that have not started, and when each last did anything.
- Which agents people actually reach forUptake is reported per agent, with runs, distinct users and last use, with the product agents people use listed apart from platform orchestration and infrastructure work.
The full answer is in How Promenaut Accounts for Every Run.
What the platform does for you, layer by layer.
Home and conversation
- What I spent today, and what my teams spent, without singling anyone out
Adoption insight
- The headline adoption measures
Kinds and roles
- Deterministic and hybrid agents
Models and routing
- A stable prompt opening can be cached
- A model call that overruns is cancelled, not abandoned
- Compare models side by side on real agent work
Conversation to work
- Cheap matching first, model call last
Execution
- Workflows refuse before they spend
- Long work does not hold an orchestrator slot
Run control
- A stop cancels the model calls already in flight
- Safe to retry a workflow start
Scheduling
- Scheduled workflows
Ingestion and provenance
- Unchanged documents are not extracted twice
Identity and access
- Separated finance permissions
- Least-privilege reporting access
Approvals, decisions and containment
- Runaway agents are suspended automatically
- Automatic step-up of containment for a runaway agent
Model risk
- One cost figure for each run and agent, with its parts by kind
Credits and budgets
- Model usage metering
- Per-run cost attribution
- Agent credit tiers
- Credit cost calculation
- Credit ledger
- What this will cost, before you run it
- A business unit out of credits stops its work
- Spend limits by person, project or business unit, per day, week or month
- Budgets allocated to business units
- An agent's cost tab: observed spend and its credit rate
- A run is charged by its agents' tiers, not its tokens
- A completed run that missed billing is billed later
- Credits move between the pool and business units as paired entries
- A warning when a business unit reaches its threshold
- The pool is funded on the contract's schedule
- Credits provisioned from a signed file
- Estimates learn from completed runs
- A module's model calls are attributed to its run
- What a run has spent, while it runs
FinOps reporting
- Spend that bought nothing
- Is this agent on the model it needs
- The optimisation surfaces need populated cost to show anything
- FinOps reporting surfaces
- Owner chargeback statement
- Renewal forecast and tier recommendation
- Credits and money reported side by side
- Per-enterprise credit currency
- Chargeback by cost centre
- The credit usage report, exportable
- Model prices held in the model registry
- Cost and capacity at a glance
Scaling and recovery
- Workflows can run on synthetic model responses
Weighing us against the tools you already know? See how we compare, with sources →
The full answers, written for you.
How Promenaut Accounts for Every RunAbout 15 minutes
Bring the numbers you need.
Cost a real workflow before it runs, and see where the spend of each run lands.
Promenaut
