Agent observability cost estimator
Estimate agent trace costs from one measured day
Agent-observability plans may charge by ingested bytes, span count, retained storage, users, queries, exports, or evaluations. This estimator converts one representative day of aggregate telemetry into the same monthly units so you can compare two plans without uploading traces, prompts, tool payloads, credentials, or customer records.
A span is one timed record inside a trace, such as a model call, tool call, or orchestration step. Measure the serialized data you actually send, enter each plan's current rate and source, and keep every unknown visible. The result is a planning estimate for a vendor conversation, not a prediction of reliability, productivity, or total operating cost.
Start with one measured dayPrivacy: Your entries should remain in this browser. Estimator inputs and results are not sent to BaristaLabs analytics, a server, or a third party. A page refresh clears the entered values.
Aggregate only: Use counts, sizes, rates, and volumes. This page does not accept files or raw telemetry.
Use a representative day and record what you measured
Choose a day that includes normal traffic, expected peaks, and the agent workflows that materially affect trace volume. A launch incident, load test, or quiet weekend is useful only when that condition is the budget baseline. Record the measurement date, environment, and serialized format so another reviewer can understand where the numbers came from.
Average span size is total serialized span bytes divided by total spans. It drives the conversion between span count and byte volume. p95 span size is the size at or below which 95 percent of measured spans fall. Use p95 to expose a large-payload tail; do not add it to the billing total or substitute it for the average.
Choose one input path. The active path is named in the result, and directly measured bytes are never added to bytes derived from span count. Decimal units are used throughout: 1 KB is 1,000 bytes and 1 GB is 1,000,000 KB.
Normalized monthly units
Enter 1.0 for no growth or headroom.
- Measured day
- Unknown
- Unknown — span count required
- Unknown — span count required
- p95: Unknown
- Projected month
- Unknown
- Unknown
- Unknown — span count required
Decimal units: 1 KB = 1,000 bytes; 1 GB = 1,000,000 KB.
Check the telemetry boundary
- Omit credentials, tokens, payment details, and unneeded fields.
- Redact customer records, private communications, regulated data, and sensitive content.
- Hash or tokenize stable identifiers when correlation is needed.
- Shorten retention for sensitive payloads and error details.
- Confirm the billing measurement after field controls change serialized volume.
Plan A / Plan B
Enter each plan in the units it actually bills. Blank means Unknown; enter 0 for an explicit zero.
Plan A
Pricing source/date unknown
Use an ISO code such as USD or EUR.
Plan A — retention
Plan A — other monthly charges
Plan A — evaluation
Deterministic checks
Model judge
Minimum commitment
This is not added as another fee.
Plan B
Pricing source/date unknown
Use an ISO code such as USD or EUR.
Plan B — retention
Plan B — other monthly charges
Plan B — evaluation
Deterministic checks
Model judge
Minimum commitment
This is not added as another fee.
Measured day → Normalized monthly units → Plan A / Plan B
Agent observability cost comparison
Planning estimate generated from aggregate inputs; no raw traces were uploaded.
Canonical page: https://www.baristalabs.io/learn/agent-observability-cost-estimator
Measured workload
- Active input path
- Path A: measured serialized gigabytes per day
- Measured decimal GB/day
- Unknown
- Average decimal KB/span
- Unknown — span count required
- p95 decimal KB/span
- Unknown — Tail indicator only; not included in billing volume.
Projected monthly units
- Billing days
- Unknown
- Traffic multiplier
- Unknown
- Projected decimal GB/month
- Unknown
- Projected spans/month
- Unknown — span count required
Plan A — monthly lines
Plan A · Currency unknown · Price per decimal GB ingested
Pricing source unknown · checked date unknown · Pricing source/date unknown
- Ingestion
- UnknownUnknown — selected rate required
- Steady-state retention estimate
- UnknownUnknown — billing rule required
- Queries
- UnknownUnknown
- Users or seats
- UnknownUnknown
- Support
- UnknownUnknown
- Export or egress
- UnknownUnknown
- Deterministic evaluation
- UnknownUnknown
- Model-judge evaluation
- UnknownUnknown
- Other fixed platform charge
- UnknownUnknown
- Known subtotal before minimum
- Currency unknown 0
- Unknown lines
- Ingestion, Steady-state retention estimate, Queries, Users or seats, Support, Export or egress, Deterministic evaluation, Model-judge evaluation, Other fixed platform charge
- Minimum commitment and credit rule
- UnknownUnknown — minimum and credit rule unresolved
- Estimated payable amount
- UnknownUnknown — missing lines or minimum-credit rule unresolved
Plan B — monthly lines
Plan B · Currency unknown · Price per decimal GB ingested
Pricing source unknown · checked date unknown · Pricing source/date unknown
- Ingestion
- UnknownUnknown — selected rate required
- Steady-state retention estimate
- UnknownUnknown — billing rule required
- Queries
- UnknownUnknown
- Users or seats
- UnknownUnknown
- Support
- UnknownUnknown
- Export or egress
- UnknownUnknown
- Deterministic evaluation
- UnknownUnknown
- Model-judge evaluation
- UnknownUnknown
- Other fixed platform charge
- UnknownUnknown
- Known subtotal before minimum
- Currency unknown 0
- Unknown lines
- Ingestion, Steady-state retention estimate, Queries, Users or seats, Support, Export or egress, Deterministic evaluation, Model-judge evaluation, Other fixed platform charge
- Minimum commitment and credit rule
- UnknownUnknown — minimum and credit rule unresolved
- Estimated payable amount
- UnknownUnknown — missing lines or minimum-credit rule unresolved
Plan subtraction is not shown; compare source-dated line items directly.
This estimate covers the entered telemetry charges. It does not infer engineering labor, incident response, migration, access control, compliance review, query performance, reliability, productivity, or business outcomes.
Questions to ask each vendor
Arithmetic cannot resolve a contract's meter, credit, retention, or exit rules. Send the same measured workload and ask every vendor to answer the same questions.
- Which bytes are metered: exported payload, accepted payload, indexed fields, uncompressed data, compressed data, or another quantity?
- Does compression, field filtering, redaction, sampling, or failed ingestion change billable volume?
- When do retention charges begin, what retention is included, and how is retained volume measured?
- Are UI, API, CLI, and MCP queries included? Are there query, compute, concurrency, or user minimums?
- Which support, platform, and minimum-commitment charges are mandatory, and which usage charges count toward the minimum?
- What are the export and egress rules during normal use and at contract exit?
- How are deterministic checks, model judges, stored scores, evaluator orchestration, and human review priced?
- Can the vendor return the same estimate using this measured workload, source date, retention period, and evaluation volume?
Pricing sources and boundaries
Vendor pages change. The following sources were checked on July 17, 2026. Their examples are date-stamped vendor evidence, not defaults, benchmarks, or recommendations in this estimator.
- Oodle's July 14 engineering post describes trace contents and vendor-reported product results.
- Oodle's product and pricing page showed vendor rates and an example calculator. No rate from it is prefilled here.
- Oodle's July 9 sampling and evaluation post discusses deterministic checks and targeted model judges. It does not define an observation as a span.
- OpenTelemetry's GenAI semantic conventions page points to the GenAI conventions repository. Conventions do not determine payload size, retention, or vendor pricing.
Related guidance for quality, outcomes, and workflow choice
If the workflow itself is still unsettled, use the AI workflow readiness assessment after this cost comparison to evaluate impact, effort, risk, ownership, and the next practical step.
Need help reviewing the full observability cost?
BaristaLabs can review one aggregate workload across ingestion, retention, evaluation, data handling, and the engineering work behind a managed or internal observability stack. See process automation for implementation support.
Review one observability workload — agent-observability cost reviewBring the aggregate comparison, pricing sources, checked dates, and unresolved vendor questions. Do not send raw traces.