New Maple Local — the whole platform as one binary.

Open-source observability.
Built for you. Native to AI.

Traces, logs, and metrics over OpenTelemetry, powered by ClickHouse. Sub-second queries across billions of rows — or let your AI agent run them for you.

Start free trial View on GitHub
1.6K
on GitHub
41.5B+
traces ingested / month
See every service at a glance.
The Maple overview dashboard — request volume, error rate and latency across every service

Trusted by teams at

HazelSakuraGymscoreSuperwallHypen

01 · Signals

Read the trace.

Every request is a span tree with its attributes intact. Hover a row; see what was on it.

An 18-span API trace waterfall with nested application and SQL operations

Search the logs.

Structured logs straight from OTLP. Severity, service, message, duration — searchable in seconds.

Structured log search with severity facets and attribute filters

Replay the session.

Every click, route, console line, and failed request. Replay and spans share one session id.

A browser session replay showing the recorded page and captured network requests

02 · Incident

From a 3 AM page to the line that caused it.

Four surfaces, one trace id. Nothing to stitch, no second tool, no tab you forgot you had open.

  1. alert trace 4f9ac21e…8b7d40f2

    The alert arrives with context.

    Error rate on api-gateway crosses 5%. The alert carries the service, the threshold it broke, and the sample traces — not a number and a shrug. Route it to Slack, Discord, PagerDuty, or any webhook.

    An alert rule with its threshold, evaluation window and firing history
  2. trace trace 4f9ac21e…8b7d40f2

    The trace opens on the failing span.

    The failing request is a span tree with every attribute intact. Three Stripe retries sit at the bottom, red, each timing out at exactly 1.75 seconds. No sampling gap, no aggregate hiding the outlier.

    A span selected in the waterfall with its full attribute set open alongside
  3. logs trace 4f9ac21e…8b7d40f2

    The logs are already correlated.

    Jump from the failing span to its logs on the trace id. Retry exhaustion, connection pool at 20/20 — the confirmation, not a second search in a second product.

    A log detail with linked trace and span IDs, correlated events and full resource attributes
  4. agent trace 4f9ac21e…8b7d40f2

    Or hand all three steps to an agent.

    Maple ships a first-class MCP server. Point Claude, Cursor, or any MCP client at it: it lists services, searches traces, finds the errors, reads the source, and proposes the fix.

    An MCP client walking the same incident — five tool calls from list_services to propose_fix, ending in a patch

03 · Platform

Everything else you'd have bought separately.

Metrics, service maps, error grouping, alerting, and an MCP server. One platform, one instrumentation, one bill.

01

Metrics & dashboards

Request rate, error rate, latency percentiles. Drag the widgets yourself, or let the agent suggest the board.

The metric explorer listing every metric with its type and cardinality
Metrics & Dashboards →
02

See dependencies move.

Live request flow across your services. The cascade you'd otherwise reconstruct from a postmortem, in the foreground.

A focused service neighborhood showing API, database, and cache dependencies with live request flow
Service Catalog & Map →
03

Error tracking

Errors grouped by type across every service, with trends over time, affected services, and sample traces attached.

Errors grouped by type with counts, affected services and last-seen times
Error Tracking →
04

Alerts with the context attached

Seven signal types with severity, incident tracking, and auto-resolution. Slack, Discord, PagerDuty, or any webhook.

Alert rules with severity, signal type and current state
05

AI & MCP

Any MCP-compatible client can list services, search traces, read the source, and propose a fix. Open protocol, no plugin.

The MCP server page with endpoint and client setup
AI & MCP Integration →
06

Connect your cloud

Kubernetes ships as a Helm chart. PlanetScale, Cloudflare, and the rest connect in one OAuth click.

Kubernetes Monitoring →
04 · Sovereignty

Your data, your instrumentation, your bill.

Maple is the observability backend you'd build if you weren't optimizing for the invoice. OpenTelemetry in. ClickHouse speed out. The source on GitHub the whole time.

01 · Open source
Source on GitHub under FSL-1.1 — each release becomes Apache 2.0 after two years. Read every line, fork it, self-host on your own boxes if that's what your security review needs.
02 · OpenTelemetry native
OTLP straight in. No proprietary agent. The instrumentation you write today moves to anywhere OTel-compatible tomorrow.
03 · Your perimeter
Run Maple on your own infrastructure, against a ClickHouse you operate and scale yourself. Your telemetry never has to leave your network — retention and residency are your call, not a fixed SaaS window.
04 · Honest pricing
Pay $39/month for 100 GB per signal, then a flat $0.30/GB. No per-host fees. No per-seat fees. The pricing page and your invoice show the same numbers.
Read the source Why OpenTelemetry See the calculator

05 · Local

The whole platform, as one binary.

One Bun-compiled maple executable — OTLP ingest, an embedded ClickHouse, a query API, and the dashboard — running on 127.0.0.1. No account, no cloud, no rate limits.

$ brew install Makisuo/tap/maple

or curl -fsSL https://maple.dev/cli/install | sh

Read the docs → View on GitHub 1.6K
local · 127.0.0.1:4318 LIVE
maple start
🍁 maple · local mode
listening on http://127.0.0.1:4318
OTLP/HTTP POST /v1/{traces,logs,metrics}
query POST /local/query
dashboard https://local.maple.dev
· bundled UI: pass --offline
data ~/.maple/data
pid 48213 · stop with maple stop
one binary · one port nothing leaves 127.0.0.1

06 · Kubernetes

The pod underneath the slow span.

Drop a Helm chart on your cluster. Kubelet, host, and kube-state metrics start flowing in. Every span your apps emit arrives carrying pod, node, and namespace, so a slow request lands you on the exact replica that served it.

Helm chart, not a black box
maple-k8s-infra is a small Helm chart you can read end-to-end.
Pod and span, joined
Spans arrive carrying k8s.pod.name, k8s.node.name, and k8s.namespace.name.
Kube-state metrics
Deployments, StatefulSets, DaemonSets, replica counts, and pod phase, all scraped via the standard kube-state-metrics exporter. 
$ helm install maple-k8s-infra maple/k8s-infra
Kubernetes monitoring → Read the Helm chart
pod heatmap · 192 pods · cluster cpu 44% LIVE
PODS 192
RUNNING 184
DEGRADED 6
FAILING 2
HOT >78% 11
default default 72
observability obs 32
kube-system kube-sys 40
ingress-nginx ingress 24
data data 24
CPU
0% 100% degraded failing

07 · Agents

Your AI agent reads it too.

Maple ships with a first-class MCP server. Any compatible agent can list services, search traces, find errors, and propose fixes. The transcript on the right walks through the exact tool calls an agent makes.

Reads the code, not just the trace
The agent pulls the source file behind a span, so the fix it proposes cites your actual code.
Writes back
Claim an issue, set its severity, attach a fix. Not a read-only dashboard.
OAuth, no plugin
Point any MCP client at the endpoint and sign in. Open protocol, nothing to install.
AI & MCP Integration
mcp · agent session LIVE
> ask maple
list services that errored in the last 15m
→ list_services { window: "15m", filter: "errors > 0" }
← 3 services
paymentserrors142p992.4s recoserrors38p99840ms shippingerrors6p99320ms
> investigate payments
→ find_errors { service: "payments", window: "15m" }
← 1 issue
stripe.charge declined  ·  142 events  ·  98 users
→ error_detail { issue_id: "iss_8a91" }
← root cause: idempotency key collision
sample trace: trace_9f3c · order 8421
→ propose_fix { issue: "iss_8a91", fix: "scope idempotency_key by user_id" }
← fix attached. opened pr#412.
> 
4 tool calls · 4.2s https://api.maple.dev/mcp
08 · The bill

The bill, line by line.

Side by side, at list price. The questions your finance team will ask, answered before they have to ask.

Line item Maple Datadog New Relic Grafana Cloud
Per-host fee None $15+ / host / mo Bundled in plan Bundled in plan
Per-seat fee None Enterprise tier $99 / full-user / mo $8+ / active-user / mo
Ingest pricing $0.30 / GB ingested $1.27 / M events $0.30 / GB $0.50 / M lines
Default retention 30d default · custom 15d default 30d default 14d default
OpenTelemetry support Native Partial (custom agent) Yes Yes
License FSL-1.1 → Apache 2.0 Proprietary Proprietary AGPL components
Self-host option Supported No No OSS components
MCP / agent surface First-class Yes Yes Yes

Competitor cells summarize public list prices — check each vendor's pricing page for current numbers. Maple's rate is the published flat per-GB. Use the calculator to project your monthly cost.

Pricing calculator → Full vs Datadog Full vs Grafana Cloud Full vs New Relic Full vs Dash0

09 · Pricing

Priced the way you'd price it.

No per-host fees. No per-seat fees. The pricing page and your invoice show the same numbers.

Startup 14-day trial
$39 /month

Everything is included.

$0
per host
$0
per seat
$0
per query
Start 14-day free trial

Free for 14 days · Cancel anytime · Card required to start

Included every month
Logs
100 GB
then $0.30 / GB
Traces
100 GB
then $0.30 / GB
Metrics
100 GB
then $0.30 / GB
Browser Sessions
5,000
then $0.002 / session

30-day retention · Private Channel support · Unlimited dashboards · Advanced alerting · MCP server · AI chat · AI error triaging · Full API access

Estimate your bill →
Enterprise

Higher volume, custom retention, priority support.

From $2,000 Talk to Founder →

Works with every language OpenTelemetry supports

FAQ

Frequently asked questions

What is Maple?
Maple is an observability platform for traces, logs, and metrics, built on OpenTelemetry and backed by ClickHouse. It lets you collect, visualize, and query telemetry from distributed systems in real time — or let an AI agent do it for you over MCP.
Is Maple open source?
Maple's source is on GitHub under the Functional Source License (FSL-1.1) — you can read every line, fork it, and self-host. Each release becomes Apache 2.0 two years after publication. You own your data and your instrumentation, and you can run Maple yourself or use the hosted version.
Is Maple OpenTelemetry-native?
Yes. Maple is built on OpenTelemetry — the open, vendor-neutral standard for telemetry — so there are no proprietary agents and no lock-in. If you already emit OpenTelemetry data, you can point it at Maple without re-instrumenting your code.
How is Maple priced?
The Startup plan is $39/month with 100 GB of logs, traces, and metrics each included, then a flat $0.30 per GB. No per-host or per-seat fees. Enterprise plans add custom volume and retention.
Does Maple work with AI agents?
Yes. Maple ships a first-class MCP (Model Context Protocol) server, so compatible AI agents can list services, search traces, find errors, and propose fixes directly against your telemetry.
Can I self-host Maple?
Yes. Maple's source is available under FSL-1.1 (each release becomes Apache 2.0 after two years), and the project can be self-hosted on your own infrastructure. The cloud plan is available for teams that prefer a managed experience.

Run it yourself.

14-day free trial, card required, no charge until it ends. Self-hosting is supported, and `maple start` needs no account at all.

Start free trial Read the source