Every endpoint, checked.
For each of 11 vendors, what Valistry does with the billing, usage and admin API. We read money, names and limits; we leave out speed, traces, content and anything that changes a setting.
Anthropic
141 endpoints in the vendor's own reference, read Sep 30, 2026: 8 read, 108 left out by rule, 25 not read yet.
Read
- Enterprise: organization cost over time by product, model and cost type, including API keys
- Enterprise: cost per seat user per model/product
- Enterprise: tokens per seat user
- Enterprise: per-user daily activity counts and role group
- Cost in dollars per day by description (model, cost_type, token_type) and workspace; no per-key split
- Claude Code per person or key per day: sessions, lines, commits, PRs, tokens and estimated cost by model
- Tokens per bucket by model, key, workspace, service account, tier, geo; web search count
- Organization members: id, email, name, role
Left out by rule
Who changed what in the vendor's console: not spend3
- Audit activity feed
- Effective organization settings
- Whether Compliance API is on
Prompts, responses or files: never read21
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
- Chats, messages, files, projects, sessions or artifacts content
It changes the vendor's settings: we only read54
- Deletes user content
- Deletes user content
- Deletes user content
- Deletes user content
- Deletes user content
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
One item of a list: we read the whole list11
- Compliance API: one group
- One API key
- One invite
- One role group
- One service account
- One increase request
- One spend limit: amount (minor units), currency, period, scope
- One member
- One workspace
- One workspace member
- One workspace service account
A setting that is not spend, a name or a limit19
- Roles
- One role
- Role permissions
- Customer-managed key configs, no money
- One external key config
- Workload identity issuers
- One issuer
- Federation rules
- One rule
- Rule workspaces
- Role catalog, no money
- One role
- Role permissions
- MCP tunnels (deprecated)
- One tunnel
- Tunnel certificates
- One tunnel certificate
- User profiles (beta), not billing
- One user profile
AWS for Bedrock
161 endpoints in the vendor's own reference, read Sep 30, 2026: 19 read, 10 covered another way, 67 left out by rule, 65 not read yet.
Read
- The bill, by day, service, usage type and tag
- Cost-allocation tag keys/values on Bedrock rows
- All budgets: limit, filters, actual and forecast spend (IAM: budgets:ViewBudget); AI-service budgets shown beside ours
- Series that exist per namespace/metric (model ids, projects, guardrails)
- Daily totals for the metrics below
- Input tokens per ModelId
- Output tokens per ModelId
- Successful calls per ModelId
- Cache read tokens per ModelId
- Cache write tokens per ModelId
- Images per ModelId, ImageSize, BucketedStepSize
- Bedrock-mantle input tokens by Project, Model
- Bedrock-mantle output tokens by Project, Model
- Bedrock-mantle completed calls by Project, Model
- Text units per guardrail and policy type: the billed unit
- One profile
- Whether model calls are logged, on or off, for Compliance; never the logs or where they go
- Tags on profiles and provisioned throughput: team keys
- One-hour keys for a connected role
Covered another way
Valistry forecasts and explains change from the ledger itself4
- AWS's own cost forecast; a comparison, not a fact
- AWS's own usage forecast
- Month-over-month cost deltas
- Drivers of a month-over-month change
AWS's own advice to buy: Valistry's opportunities come from your runs6
- AWS purchase recommendation
- Recommendation detail
- RI recommendation
- EC2 rightsizing; GPU instances
- Result of a commitment analysis
- Commitment analyses list
Left out by rule
A setting that is not spend, a name or a limit12
- Tag backfill history
- Generation job status
- Tags on CE resources
- Tags on exports
- Tags on budgets
- Copy jobs
- Import jobs
- Skip
- Policies
- Retention setting
- Model access form
- Resource policy
A per-minute series: the day's total is what the ledger keeps10
- Record counts for sizing resource-level data
- Older one-metric read; GetMetricData covers it
- Estimated TPM quota use per minute
- Errors
- Errors
- Throttles
- Invocation-log delivery health
- Per-inference datum for percentiles
- Per-inference datum
- Errors
It changes the vendor's settings: we only read40
- Mutates Cost Explorer config
- Mutates Cost Explorer config
- Mutates Cost Explorer config
- Mutates Cost Explorer config
- Mutates Cost Explorer config
- Mutates Cost Explorer config
- Mutates Cost Explorer config
- Mutates Cost Explorer config
- Mutates Cost Explorer config
- Mutates Cost Explorer config
- Mutates Cost Explorer config
- Mutates Cost Explorer config
- Mutates Cost Explorer config
- Mutates Cost Explorer config
- Mutates Cost Explorer config
- Mutates Cost Explorer config
- Mutates exports
- Mutates exports
- Mutates exports
- Mutates exports
- Mutates exports
- Mutates legacy CUR
- Mutates legacy CUR
- Mutates legacy CUR
- Mutates budgets
- Mutates budgets
- Mutates budgets
- Mutates budgets
- Mutates budgets
- Mutates budgets
- Mutates budgets
- Mutates budgets
- Mutates budgets
- Mutates budgets
- Mutates budgets
- Mutates budgets
- Mutates budgets
- Mutates budgets
- Mutates budgets
- Every mutating Bedrock control-plane action
Who changed what in the vendor's console: not spend1
- Action history
Speed, not money: observability, which we leave to the tools built for it4
- Skip
- Skip
- Guardrail calls, errors, latency
- Skip
Azure for OpenAI and AI services
70 endpoints in the vendor's own reference, read Oct 23, 2026: 7 read, 14 covered another way, 26 left out by rule, 23 not read yet.
Read
- The same cost query at a billing account: every subscription with cost, each filtered to itself
- The same cost query at a management group
- Your tag keys on a subscription's cost, to find the ones that name owners
- AI services accounts in the subscription: name, kind, region
- Cost per day by meter, for OpenAI, partner models in Marketplace and every other AI service; actual, amortized beside, and by your tags
- Tokens per model deployment per day: prompt, generated, cached
- Subscriptions the app can see, each one read
Covered another way
The same cost per meter per day, read from Cost Management's query3
- Cost summed over a management group
- Charges per billing period
- Usage and cost per meter per day, older API
The same token counts, read per account from the metrics endpoint2
- The same metrics for every resource of a type in a region
- The same metrics for many resources in one call
The same accounts, read for the whole subscription1
- The same accounts, one resource group
The vendor's own advice to buy: Valistry's opportunities come from your runs3
- Azure's advice on savings plans to buy
- Detail of one reservation recommendation
- Azure's advice on reservations to buy
Partner models read from the cost query by publisher type, with the offer and the meter1
- Marketplace charges, such as partner models sold through Azure
The same tag keys, read per subscription1
- Tag keys and values on cost records
Valistry forecasts and explains change from the ledger itself1
- Azure's own cost forecast; a comparison, not a fact
Cost split by your tags, and amortized beside, read from the query1
- Line-by-line cost file with tags, actual or amortized
The same cost per meter per day, read from the query1
- Older detailed cost file
Left out by rule
A setting that is not spend, a name or a limit16
- Operations the service offers
- Deleted accounts that can still be restored
- Commitment tiers on offer: a price list, not your spend
- Models offered in a region
- Where capacity is free for a model
- Account types and tiers on offer
- The account's keys: secrets, never read
- Models this account can deploy
- Content filter policies
- Which metrics a resource has
- Metric namespaces of a resource
- Scheduled cost exports to a storage account
- When an export ran and whether it worked
- Scheduled cost emails and anomaly alerts
- Cost Management settings for a scope
- Saved cost views
One item of a list: we read the whole list5
- One account
- One deployment
- One budget
- The file one cost details report produced
- Status of one detailed cost report
It changes the vendor's settings: we only read5
- Creates or changes an account
- Creates or changes a deployment
- Creates or changes a budget
- Creates or changes an export
- Runs an export now
GitHub Copilot
59 endpoints in the vendor's own reference, read Sep 30, 2026: 6 read, 1 covered another way, 21 left out by rule, 31 not read yet.
Read
- The organization the Valistry app is installed on
- Repositories the Valistry app was given
- Copilot plan, seat counts this cycle, and policies
- Each seat: holder, team, last activity and editor
- Repositories, newest push first
- Closed pull requests: merged or not, when, and the branch
Covered another way
The same merged pull requests, read per repository1
- Pull requests found by search
Left out by rule
It changes the vendor's settings: we only read12
- Sets the enterprise cloud agent policy
- Creates an enterprise budget
- Creates a cost center
- Adds people or repositories to a cost center
- Creates a budget
- Deletes a budget
- Changes a budget
- Cancels teams' seats
- Gives teams Copilot seats
- Cancels people's seats
- Buys seats for people
- Changes where the cloud agent may work
One item of a list: we read the whole list5
- One enterprise budget
- One cost center
- One budget
- One member's seat
- One pull request
A setting that is not spend, a name or a limit1
- Which repositories Copilot's cloud agent may work in
Prompts, responses or files: never read3
- Commits and their messages
- Changed files and their diffs
- Review comments
Cursor
69 endpoints in the vendor's own reference, read Sep 30, 2026: 3 read, 3 covered another way, 37 left out by rule, 26 not read yet.
Read
- Per person per day: requests, lines, model most used, active or not
- Each request: person, model, kind, tokens, cents charged beyond the plan
- Team members: email, name, role, removed or not
Covered another way
The same use per request, read from filtered usage events2
- Model use per day per person
- Model use per day across the team
Who was active each day, read from daily usage data1
- Daily active people
Left out by rule
Prompts, responses or files: never read5
- Each accepted AI change
- The same changes, as a file
- One commit with line-by-line blame
- Each Bugbot review and its findings
- Insights drawn from conversations
A setting that is not spend, a name or a limit10
- Grok Bot access
- Grok Bot auto review
- Grok Bot capabilities
- Grok Bot network
- Grok Bot setup manifests
- Grok Bot team rules
- Repositories Cursor must not index
- Which models the team may use (preview)
- Providers and models, with access settings
- Models of one provider
It changes the vendor's settings: we only read19
- Turns Grok Bot on or off, and changes its rules
- Creates or replaces a blocklist
- Removes a blocklist
- Creates a directory group
- Deletes an empty group
- Renames a group or changes its limits
- Adds members to a group
- Removes members from a group
- Creates a billing group
- Deletes a billing group
- Changes a billing group
- Removes members from a billing group
- Adds members to a billing group
- Changes the model policy
- Turns a provider on or off
- Turns a model on or off
- Removes a member
- Sets one person's spend limit
- Sets spend limits for up to 100 people
Who changed what in the vendor's console: not spend1
- Audit log of admin changes
One item of a list: we read the whole list2
- One directory group
- One billing group and its members
Databricks
48 endpoints in the vendor's own reference, read Sep 30, 2026: 4 read, 2 covered another way, 26 left out by rule, 16 not read yet.
Read
- Runs one read query on the billing system tables
- That query's result, polled until done
- List price of each SKU over time
- Billable use per day: SKU, workspace, DBUs, tags, who ran it
Covered another way
The same use, read from system.billing.usage1
- Billable use as a file, per workspace and SKU
The outside provider bills it; read from that provider's connector1
- Databricks' estimate of spend on outside models through its gateway
Left out by rule
A setting that is not spend, a name or a limit3
- The account's usage dashboard
- Where logs are delivered
- The endpoint's request schema
It changes the vendor's settings: we only read12
- Creates a usage dashboard
- Creates a budget policy
- Creates a budget
- Changes a budget
- Creates a serving endpoint
- Creates a provisioned throughput endpoint
- Deletes an endpoint
- Changes gateway settings: limits, logging, guardrails
- Changes an endpoint
- Changes an endpoint's rate limits
- Changes an endpoint's tags
- Cancels a query
One item of a list: we read the whole list4
- One budget policy
- One budget
- One serving endpoint
- One chunk of a large result
A per-minute series: the day's total is what the ledger keeps2
- Endpoint metrics: requests, latency, errors
- Machine use by the minute
Traces of single calls: observability, and they can carry content2
- A served model's build logs
- A served model's recent logs
Prompts, responses or files: never read2
- Messages sent to the assistant
- Every query run, with its text
Who changed what in the vendor's console: not spend1
- Audit events from every workspace
LiteLLM
122 endpoints in the vendor's own reference, read Sep 30, 2026: 4 read, 25 covered another way, 66 left out by rule, 27 not read yet.
Read
- Each request: spend, tokens, model, key, team and person. If your proxy stores prompts, we drop them before saving
- Fallback for proxies without v2, with summarize=false; per-request rows (same content caveat)
- Keys: hashed token, alias, team, user, organization, spend, max_budget, tpm/rpm limits, expiry
- Teams: alias, organization, members, spend, max_budget, budget_duration, tpm/rpm, models
Covered another way
The same money as the spend log we read, summed25
- Spend per request tag over a date range (Enterprise)
- Spend per key (all-time)
- Spend per internal user
- Computes cost for a model and token counts from the proxy price map; not a record of spend
- Total spend and max budget for the proxy
- Daily spend grouped by team, customer or api_key, with model breakdown (Enterprise)
- Daily spend totals, optionally for one key
- Top keys by spend
- Spend per team per day
- Top models by spend
- Spend per tag over a date range (Enterprise)
- Spend per provider
- Top end users by spend
- Spend report for keys over a date range
- Spend report per internal user
- Spend report per team
- Spend report per organization
- Daily spend, tokens, requests by model, provider, key (pre-aggregated table); cheaper than raw logs
- Same, summed over the range
- Daily spend and tokens per team, by model and key
- Same, summed over the range
- Team spend split by member
- Daily spend and tokens per organization
- Daily spend and tokens per tag
- Daily spend and tokens per customer (end user); also /end_user/daily/activity
Left out by rule
Prompts, responses or files: never read2
- Request and response body for one call
- Session log view for the UI
A per-minute series: the day's total is what the ledger keeps5
- Share of requests whose spend was logged; a data-quality check worth one read
- Requests and tokens per day, UI chart
- Requests and tokens per model per day
- Error counts
- Error counts per deployment
It changes the vendor's settings: we only read57
- Resets spend counters
- Refreshes spend views
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
- Mutates proxy state
A setting that is not spend, a name or a limit2
- Team metadata schema
- Team member permissions
OpenAI
112 endpoints in the vendor's own reference, read Sep 30, 2026: 5 read, 8 covered another way, 82 left out by rule, 17 not read yet.
Read
- Cost in dollars, per day by line_item and project (also api_key_id); the bill, including fine-tuning and storage
- Data retention setting, for Compliance
- Transcription seconds by model, project (read to join minutes to bill lines)
- Tokens (input, cached, output, audio) and requests by model, project, key and person (kept as a hash), batch, service_tier. Read by model+project+key
- Fine-tuning runs: base model and trained tokens (needs a project key)
Covered another way
Priced from each line of the bill, which we read8
- TTS characters and requests by model, project, key
- Code interpreter/container sessions by project
- Input tokens and requests by model, project, key, user
- File search tool calls by project (and vector store)
- Images by model, size, source (generation/edit/variation), project, key
- Input tokens and requests by model, project, key, user
- Vector store bytes by project; storage the bill charges per GB-day
- Web search tool calls by model, project, key
Left out by rule
It changes the vendor's settings: we only read57
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
- Mutates organization state
One item of a list: we read the whole list12
- One admin key
- One group
- One group membership
- One invite
- One project
- One project key
- One project group
- One service account
- One project spend alert
- One project member
- One organization spend alert
- One member
Who changed what in the vendor's console: not spend1
- Audit event feed
A setting that is not spend, a name or a limit12
- MTLS certificates, no money
- One certificate
- Roles on a group
- One group role
- Project certificates
- Project data retention setting
- Custom role catalog
- One role
- A user's custom roles
- One user role
- Customer-managed storage configs
- One storage config
OpenRouter
49 endpoints in the vendor's own reference, read Sep 30, 2026: 1 read, 1 covered another way, 36 left out by rule, 11 not read yet.
Read
- Spend and tokens per day, model and provider, last 30 days
Covered another way
A management key spends nothing; spend is read from activity1
- The calling key's own spend and limit
Left out by rule
A setting that is not spend, a name or a limit16
- What the analytics query can group and filter by
- Batch jobs and their state
- Your own provider keys used through OpenRouter
- Public daily token totals of top models
- Public cost per session across apps, not your spend
- Embedding models
- Endpoints that keep no data
- Model catalog with list prices
- How many models exist
- Models your settings allow
- Providers serving one model, with their prices
- Where OpenRouter sends traces
- Saved request presets
- Private endpoints and their prices
- Upstream providers and their policies
- How directory groups map to workspaces
Prompts, responses or files: never read2
- Uploaded files
- One request's prompt and response
Traces of single calls: observability, and they can carry content1
- One request's tokens, cost and timing
It changes the vendor's settings: we only read10
- Sends feedback on a request
- Creates a guardrail, and assigns or removes keys and members
- Creates a key
- Deletes a key
- Changes a key's name, limit or state
- Starts a directory sync
- Creates a workspace
- Changes a workspace
- Adds members to a workspace
- Sets a workspace budget
One item of a list: we read the whole list7
- One guardrail
- Keys under one guardrail
- Members under one guardrail
- One key
- One model
- One workspace
- One workspace budget
Snowflake Cortex
40 endpoints in the vendor's own reference, read Sep 30, 2026: 3 read, 5 covered another way, 5 left out by rule, 27 not read yet.
Read
- AI functions per hour: function, model, credits, tokens, pages
- Runs one read query on the Cortex usage view
- That query's result, polled and read part by part
Covered another way
The same AI functions, read from CORTEX_AI_FUNCTIONS_USAGE_HISTORY3
- AI SQL function credits, earlier name
- AI function credits per query
- Older view of AI function credits per hour
Valistry forecasts and explains change from the ledger itself1
- Snowflake's own spend anomalies
The same view, read in each connected account1
- AI functions across every account in the organization
Left out by rule
It changes the vendor's settings: we only read2
- Changes a budget's limit
- Cancels a query
Who changed what in the vendor's console: not spend1
- Sign-ins
A per-minute series: the day's total is what the ledger keeps1
- Credits per hour by service type
Prompts, responses or files: never read1
- Every query run, with its text
Google Cloud for Vertex AI
42 endpoints in the vendor's own reference, read Sep 30, 2026: 2 read, 4 covered another way, 23 left out by rule, 13 not read yet.
Read
- Standard billing export: cost per SKU, project and day, with credits
- One read query on your billing export, for Vertex and other AI
Covered another way
The billing export names each project on its rows3
- Projects paid by a billing account
- Which billing account pays for a project
- Projects by name
Priced from each line of the bill, which we read1
- List prices per SKU, Vertex models included
Left out by rule
A setting that is not spend, a name or a limit7
- Datasets in a project
- Tables in a dataset
- Who can see or pay the billing account
- Google Cloud services in the price catalog
- Quota increases asked for
- Which metrics exist
- Model Garden catalog
It changes the vendor's settings: we only read9
- Starts a job, which can write tables
- Creates a budget
- Changes a budget
- Creates a billing account
- Renames a billing account
- Creates a sub-account
- Moves a billing account
- Changes who can use the billing account
- Changes which account pays for a project
One item of a list: we read the whole list4
- The rest of one query's result
- One budget
- One billing account
- One endpoint
A per-minute series: the day's total is what the ledger keeps2
- Vertex tokens, requests and latency per model, by the minute
- The same metrics, by query
Prompts, responses or files: never read1
- Cached prompt content