Open SourceNow accepting design partners

Nobody touches the pipeline.
So the bill keeps climbing.

  • Logs nobody reads. Dashboards nobody trusts.
  • One person owns the config — everyone else is afraid to touch it.
  • Last time someone did, monitoring went dark mid-incident.
  • So nothing changes. And the invoices climb.

Magpie makes OTel pipeline changes safe — canaried, health-gated, auto-rolled-back. Ship config changes without the fear.

Looking for teams with 50+ collectors under cost pressure. No credit card required.

~90%
of telemetry never queried
< 2s
safe fleet-wide config push
0
incidents from config changes
Works with any OTLP-compatible backend
GrafanaDatadogHoneycombSplunkNew Relic
The problem

Your pipeline is too fragile to touch.

Infra cost keeps climbing. Nobody wants to touch the pipeline because it might break production. So nothing changes — and monitoring breaks during incidents anyway because the config drifted months ago.

90%

of collected telemetry is never queried

Industry research
7–12%

of cloud spend goes to observability

Platform engineering surveys
96%

of orgs actively controlling observability costs

2026 reports

Has your monitoring ever gone dark because the pipeline was too fragile to touch?

Debug logs shipping to prod. Duplicated exporters nobody remembers adding. Metrics no dashboard references. Everyone knows it's there — but one bad config push means 40 minutes of blindness during an incident. Remove the blast radius and the savings unlock themselves.

How savings deploy safely

Ship cost cuts without breaking observability.

Magpie identifies what to cut — then deploys the fix through a phased rollout. Canary first. Health-gated promotion. Automatic rollback if anything degrades. Your observability stays intact while your bill drops.

Cost Optimization #12
est. saving: $8,200/mo
Deploying safely
Validate
Canary
Soak
Promote
Saving
Fleet Status
Savings applied0/200
Deploying0/200
Rolled back0/200
Observability Health
Pipeline throughput
stable
Error visibility
100%
Alert coverage
no gaps
no observability regression detected
What's Being Cut
-debug logs (never queried)
-info logs (no dashboard ref)
+filter: severity >= WARN
~340 GB/day log reduction
soak: 4:32 remaining

Savings are reversible

Every cost-cutting config change auto-rolls-back if health degrades. Drop debug logs today, undo in 2 seconds if something needs them.

Canary before fleet

A filter that saves $8K/mo hits 10 hosts first. If pipeline throughput dips or error rates spike, the rollout halts — your fleet never sees it.

See the blast radius

Before you drop a log level or sample a trace pipeline, Magpie shows exactly how many hosts and how much volume the change affects.

No observability gaps

Health signals are watched continuously during soak. If the savings come at the cost of visibility, the system catches it before you do.

Audit everything

Every cost optimization applied, reverted, or promoted is hash-chained in an append-only log. Know what changed, when, and who triggered it.

Git stays source of truth

Every applied recommendation writes back as a PR. Your team reviews cost changes like any other code change. No governance regression.

The reason teams don't cut telemetry costs isn't that they can't find what to cut — it's that deploying the fix is terrifying. Magpie makes it safe, so the savings actually happen.

Free forever — Apache 2.0, unlimited agents, no feature gates on safety
Cost optimizationEarly Access

Cost control, through your own fleet.

Magpie analyzes collector self-telemetry — metadata only, never your actual data — to find what to cut and ship the fix. Each stage earns the trust for the next.

0
Stage 0
See

Per-pipeline volume and cardinality accounting from collector self-telemetry. Know what you ship, where, and what it costs.

Building with design partners
1
Stage 1
Suggest

Concrete config diffs that drop never-queried logs, sample high-volume traces, and deduplicate exporters. With projected savings.

Building with design partners
2
Stage 2
Apply on approval

One-click apply through canary + rollback. Every recommendation ships through the safe-change engine. PR write-back to git.

Roadmap
3
Stage 3
Auto-apply

Policy-bounded autonomous optimization. "Keep log volume under X GB/day, never touch error-level." Earned via ledger history.

Roadmap

No new data path

Acts through your existing OTel collectors. No proprietary agent, no proxy, no man-in-the-middle.

No ingest conflict

Unlike backend vendors, we have zero revenue tied to your telemetry volume. We win when your bill drops.

Metadata only

Analyzes throughput, cardinality, and pipeline stats. Your actual telemetry data never transits Magpie.

Why not...

The empty square nobody occupies.

Neutral. Self-hostable. Cost optimization through existing OTel collectors. With safe-change machinery. Nobody else is here.

GitOps / Ansible
"Do nothing"
Their gap

Minutes-to-hours rollouts. Zero telemetry-aware safety. Override sprawl at scale.

Magpie

Sub-2s propagation with canary gates and auto-rollback. One system for K8s and VMs.

Grafana FM / Coralogix / Bindplane
Free bundled fleet managers
Their gap

Cloud-locked. No self-host. Structurally conflicted against cutting your ingest bill.

Magpie

Self-hosted, Apache 2.0. Backend-neutral — we have no ingest revenue to protect.

Cribl / Sawmills / Grepr
Proprietary cost platforms
Their gap

New vendor agent in your data path. $2,500/mo+ floors. Not OTel-native.

Magpie

Acts through your existing otelcol-contrib. No new agent. No new data path. OTel-native.

DIY OpAMP
opamp-go + alpha supervisor
Their gap

Deliberately incomplete. Example server only. No cohorts, canary, or rollback. You build and maintain everything.

Magpie

Production-ready control plane. Phased rollouts, scoped targeting, audit trail — shipped.

Architecture

One binary. No dependencies. Five-minute deploy.

Single Go binary. SQLite for small fleets, Postgres when you scale. The agent wraps vanilla otelcol-contrib — unmodified, unforked, always upstream. Magpie is not in the data path.

No new agentYour existing otelcol-contrib. Always upstream.
Not in the data pathControl plane only. Telemetry never routes through Magpie.
Survives control-plane failureAgents keep running last-applied config indefinitely.
K8s + VMs, one system51% of deployments include VMs. Magpie handles both.
1
Your OTel Collectors
Vanilla otelcol-contrib on every host
K8s DaemonSets · VMs · Bare metal · Gateways
OpAMP (health + config hash)
2
magpied
Single-binary control plane
Rollout engine · Scope targeting · Audit log · REST API
Config push < 2s
3
Cost IntelligenceEarly Access
Metadata-only analysis
Usage analyzer · Cost detector · Recommendations · Savings ledger
Design Partners

Become a design partner.

We're looking for platform teams running 50+ OTel collectors under cost pressure. Get free Business-tier access for 6 months, direct support, and shape the cost optimization roadmap.

Looking for teams with 50+ collectors under cost pressure. No credit card required.

Free for 6 months

Full Business tier — cost intelligence, automation, priority support

Shape the product

Direct roadmap influence. Your pain points become our sprint priorities.

5-min quickstart

Apache 2.0, single binary. Deploy the wedge today, cost layer when ready.

Star on GitHub|Apache 2.0 · Self-host forever