Logfire Pricing
Logfire's paid tiers don't buy more volume. They buy the right to go over the same free-tier cap.
Logfire's real numbers, verified straight from its own pricing page: Personal is $0 a month, Team is $49, Growth is $249, and Enterprise is custom through sales. What none of the three published tiers change is the ten million logs, spans, and metrics included every month, the same allowance on Personal as on Growth. The difference is what happens once a workspace hits it, what a single AI agent conversation actually costs in billable records, and a built-in AI Gateway markup that quietly drops as a workspace upgrades. This page verifies every number and does the math no pricing page runs for you.
Verified against Logfire's own pricing page
What Logfire actually costs
Logfire's pricing runs four tiers, and every figure below comes straight from Logfire's own current pricing page. Personal costs $0 a month, includes ten million logs, spans, and metrics, and holds thirty days of retention for one admin plus two read-only guests, with community support and no overage option at all: ingestion simply pauses once the ten million allowance runs out for the month. Team costs $49 a month, keeps that same ten million included records, and adds five seats with room to buy up to twelve total at $25 a seat, ten read-only guests, and a spending cap option with a money-back guarantee, still on thirty day retention. Growth costs $249 a month, keeps the identical ten million included records a third time, stretches retention out to ninety days, removes the seat limit entirely for unlimited seats and unlimited read-only guests, adds priority support and a boilerplate HIPAA business associate agreement, and gets early access to new features before they ship broadly.
Enterprise is where the published numbers stop. Logfire's page lists retention, volume, and support all as custom, reachable only through a sales conversation, and adds what the three published tiers don't: multi-year retention options, deployment as cloud, dedicated, or fully self-hosted, a dedicated Slack channel, and 24/7 SLA-backed support. It also adds full access to Logfire's AI Gateway as an add-on, with data loss prevention controls and observability SLOs for the underlying LLM providers layered on top, priced custom rather than at a fixed rate. Logfire's own site confirms the core per-record overage rate, $2 per million logs, spans, or metrics past the included allowance, has stayed flat through a January 1, 2026 repricing that tightened seat counts and guest limits around it rather than touching that number.
Same cap, different everything else
Logfire's real structure: one shared cap, four different deals around it
Straight from Logfire's own pricing page, not a third party estimate.
Personal: $0 a month
Ten million included records, thirty day retention, one admin plus two read-only guests. No overage available: ingestion pauses once the cap hits.
Team: $49 a month
Same ten million included records, five seats (up to twelve at $25 each), ten guests, $2-per-million overage once the cap hits instead of a pause.
Growth: $249 a month
Same ten million included records again, unlimited seats and guests, ninety day retention, priority support, and a lower AI Gateway markup.
Enterprise: custom pricing
Custom volume, custom retention, self-hosted or dedicated deployment, 24/7 SLA support, and full AI Gateway access with data loss prevention. No published number anywhere.
The math Logfire's own pricing page never runs
What one real agent conversation costs in billable records
Logfire's billing documentation is unusually direct about its unit: every span, log line, or metric point shipped counts as one record, no exceptions beyond the aggregate metrics that FastAPI or SQLAlchemy auto-instrumentation roll up into a single count regardless of how many requests sit underneath them. What that plain definition hides is how fast it adds up once an AI agent, not a simple web request, is the thing being traced. A single customer turn handled by a support copilot typically generates one span for the incoming request, one for the agent invocation, one span per tool call it makes along the way, one for the model completion itself, and one or two more from framework auto-instrumentation underneath it. That lands somewhere around seven to nine billable records for one single back-and-forth, before counting anything a retrieval step or a second tool call would add.
Run that arithmetic forward and a five-message conversation that fully resolves a support ticket costs roughly forty billable records. To exhaust the ten million records every Logfire tier includes for free, a team would need something in the range of two hundred fifty thousand fully resolved conversations in a single month, north of eight thousand a day. Most teams shipping a first customer-facing agent sit well under that number, which means the free tier's real constraint is rarely the record cap itself; it's the two-guest seat limit and the thirty-day retention window that push a growing team toward Team or Growth long before the record count ever becomes the binding constraint.
What the pricing page alone won't tell you
Three things Logfire's own pricing table doesn't spell out
None of these show up comparing the four price tags side by side. All three show up once a real agent workload is running.
The free and paid tiers share one number
Ten million included logs, spans, and metrics a month on Personal, Team, and Growth alike. Upgrading buys more seats, longer retention, and the right to keep shipping data past that cap, not a bigger cap itself.
An agent trace is nearly all spans, and spans don't roll up
FastAPI and SQLAlchemy's own metrics count once no matter how many requests they summarize. A multi-step agent trace gets no such rollup: every tool call and every model completion bills as its own separate record.
Bring your own key and the tier stops mattering
Logfire never marks up a bring-your-own-key model call on any tier. Route through Logfire's own built-in provider credentials instead, and the markup drops from five percent on Personal and Team to three percent on Growth.
FAQs
Logfire pricing: frequently asked questions
Common questions from teams pricing Logfire against a real agent workload before committing to a tier.
How much does Pydantic Logfire actually cost?
Four tiers, verified straight from Logfire's own pricing page. Personal costs $0 a month, includes ten million logs, spans, and metrics, and holds thirty days of retention for one admin plus two read only guests. Team costs $49 a month, keeps the same ten million included records but adds five seats (additional seats run $25 each up to twelve total) and ten read only guests, still on thirty day retention. Growth costs $249 a month, keeps the same ten million included records again, stretches retention to ninety days, and removes the seat cap entirely: unlimited seats and unlimited read only guests. Enterprise has no published number, quoted only through Logfire's sales team, with custom retention, custom volume, and the option to self host instead of running on Logfire's own cloud.
Wait, don't the paid tiers include more volume than the free tier?
No, and this is the detail every third party review of Logfire skips past. Personal, Team, and Growth all list the identical ten million logs, spans, and metrics a month as their included volume. The free tier's limit is a hard cap: hit ten million records and ingestion pauses until the next billing cycle, with no way to pay for more. Team and Growth carry the exact same ten million included allowance, but past it, ingestion keeps running at $2 per million additional records rather than stopping. The $49 and $249 price tags are not buying a bigger bucket. They are buying the right to keep shipping data once the bucket everyone gets for free is already full, on top of more seats, longer retention, and a cheaper markup on Logfire's built in AI Gateway.
What actually counts as one billable record?
Logfire's own billing documentation states it plainly, unlike several comparable tracing tools that leave their unit undefined: every span, log line, or metric point shipped to Logfire counts as one record, full stop. A single OpenTelemetry trace from one incoming request can contain many spans, and each one bills separately rather than the whole trace counting once. Aggregate metrics are the one exception. Auto instrumentation for FastAPI or SQLAlchemy emits a rolled up metric that counts once no matter how many underlying requests or queries it summarizes, so metrics tend to stay cheap relative to spans. Spans do not get that same rollup, and an AI agent's trace is usually made of nothing but spans.
How many billable records does one real AI agent conversation actually produce?
Logfire never publishes this math, so it is worth doing directly. A single customer turn handled by a support copilot built on Pydantic AI or LangChain typically produces one span for the incoming request, one for the agent invocation itself, one span per tool call (an order lookup and a policy check is two more), one span for the model completion, and one or two more from FastAPI or SQLAlchemy auto instrumentation underneath it. That lands around seven to nine billable records per single turn. A five message conversation that fully resolves a ticket runs somewhere near forty records. At that rate, a team would need roughly two hundred fifty thousand fully resolved conversations in a single month, north of eight thousand a day, to exhaust the ten million records every tier includes for free. Most teams shipping their first customer facing agent are nowhere near that volume; the ones who are tend to already be Logfire's paying Growth or Enterprise customers for the seat count and retention window alone.
Is Logfire's AI Gateway markup the same across every tier?
No, and the gap is easy to miss reading the pricing page quickly. Bring your own API key for OpenAI, Anthropic, or another model provider through the AI Gateway and Logfire never marks it up, on any tier, from Personal through Enterprise. Route a call through one of Logfire's own built in provider credentials instead and Personal and Team both charge a five percent markup on that call, while Growth drops the same markup to three percent. Enterprise gets full AI Gateway access as an add on with data loss prevention controls and observability SLOs layered on top, priced custom rather than at a fixed percentage. A team that bring their own keys pays the identical rate at every tier; a team that leans on Logfire's built in credentials pays a real, tier dependent tax on every model call that routes through them.
How does Logfire's per-span pricing compare to Arize, Langfuse, or LangSmith?
Badly on a straight dollar comparison, cleanly on a plain language one. Logfire's $2 per million records is the easiest unit of the four to actually explain: one span, one log line, or one metric point, no exceptions worth memorizing. Arize bills against a span it never formally defines on its own pricing page. Langfuse bills against a unit built from three different event types added together, trace, observation, and score. LangSmith stacks two separate meters, compute units and storage units, that never resolve into one number without a spreadsheet. None of the four are cheap once a workload gets serious, but Logfire is the one where a team can actually predict next month's bill from this page's own math instead of guessing at an undocumented formula.
Does Logfire show anything to my own paying customers, and can I run AiAgRe alongside it?
No plan does, Personal through Enterprise, and running both together is the normal setup rather than a workaround. Every dial on Logfire's dashboard, the ten million record allowance climbing toward its cap, the retention window, the AI Gateway markup on a routed call, lives inside a workspace built for the team that owns it. Unlimited seats on Growth means unlimited seats for your own engineers, not a scoped view for an outside customer, and Enterprise's self hosting and dedicated Slack channel still answer to your company alone. Logfire's own documentation has no mention anywhere of a second tenant layer, a customer facing view, or a white labeled dashboard, because that was never the product it set out to build. AiAgRe's Node SDK reads the same underlying agent events Logfire already traces through the same Pydantic AI, LangChain, or LlamaIndex integrations, tags each one with an organization identity and a customer identity at the point of ingestion, and turns that into white-labeled dashboard components your own customer sees inside your product, showing deflection rate and cost per resolution scoped so tightly that one customer's numbers never reach another's view. Installing it changes nothing about how Logfire counts a span or bills a Gateway call; the two tools read the same agent, one billed and displayed to your team, the other billed and displayed to the customer paying for it.
Already tracing your agent with Logfire? Now show your customer what it did.
Request access and we'll walk through how AiAgRe's embed tokens sit alongside the spans Logfire already traces.