Obstack Bee

Distributed tracing and profiling. Inside the service, on the host.

Bee traces and profiles every process on a Linux host: inside a microservice, between services on the same machine, then across hosts over gossip. HTTP, gRPC, TLS, MySQL and Redis from the kernel, named Node.js awaits, no SDK and no coordinator.

See the plans Deployment guide Request an evaluation license

Example trace · checkout-api812 ms

POST /api/checkout

812 ms

redis GET cart:*

2 ms

mysql SELECT … orders

41 ms

await reserveStock()

640 ms

HTTPS POST /reserve

598 ms

code.function reserveStock · src/inventory.js:88parent proven by socket cookie · traceparent

01 / What Bee does

Trace and profile. One binary, no coordinator.

→ Trace

Distributed traces inside a microservice, between processes on the same host, and across hosts over gossip.

→ Profile

On-CPU and await stacks tied to the request that held them. Profiling is the same evidence as the trace, not a second product.

→ Correlate

Links hops and async waits only where socket, stream or runtime evidence proves it.

→ Export

Sends OTLP over HTTP to any OpenTelemetry collector and serves Prometheus metrics locally.

02 / Under the hood

Built for the engineer who reads the trace and asks how it knows.

In-kernel capture that never gets in the way

Aya cgroup-skb programs filter TCP in the kernel and keep per-CPU flow aggregates. Bee never blocks or drops a packet: filtering only decides what gets observed.

eBPF · cgroup ingress/egress · OpenSSL uprobes

Protocols parsed to the field

HTTP/1 with pipelining, HTTP/2 and gRPC with HPACK state, TLS handshakes with SNI and ALPN. Full MySQL query text with typed prepared-statement parameters. Redis RESP2 and RESP3.

HTTP/1 · HTTP/2 · gRPC · TLS 1.2/1.3 · MySQL · Redis

Node.js awaits, named

Promise probes capture a bounded stack at the exact await boundary. Bee reads V8 frames using the layout constants the Node binary itself exports, and emits a span only once an observed socket proves which request it belongs to.

Node.js ≥ 22 · process_vm_readv · v8dbg_*

Evidence, never guesses

Parent–child edges come from socket cookies, HTTP/2 stream ids, MySQL statements, W3C traceparent and authenticated hops between agents. Timestamp overlap proves nothing, so without proof a span ships as an honest root.

unmeasured values are labelled, never zero

Intra-service and intra-host

Bee follows a request through awaits, SQL and Redis inside one process, then across local sockets to other processes on the same machine. Cross-host hops continue the same trace over gossip.

intra-microservice · intra-host · gossip

Works with your collector

Batches go over one persistent OTLP/HTTP connection to any OpenTelemetry collector: Vector, the OpenTelemetry Collector or your vendor’s endpoint. If it goes down, finished batches wait up to five minutes in a fixed on-disk ring.

OTLP/HTTP protobuf · rustls · at-most-once

03 / Distributed traces

One trace: inside the service, on the host, across hosts.

Bee already follows a request through awaits and local sockets on one machine. Gossip continues that same distributed trace when it hops hosts — no coordinator to run, scale or lose. Profiling rides along: the stack that held the request, not a dump from somewhere else.

No coordinator

Membership, liveness and peer addresses live on the agents. There is no extra process to deploy, scale or lose when a request hops machines.

Gossip between Bees

Each node talks to a few peers. Who is up, how to reach them, and enough identity to continue a trace travel with the gossip — not through a central correlator.

One trace across hosts

When a request leaves one Linux box for another, Bee authenticates the hop and continues the W3C trace. The next agent already knows who sent it.

On your network only

Gossip stays inside the fleet. Nothing phones home to stitch traces together. Air-gapped delivery and offline updates are quoted as Enterprise, not on the list plans.

04 / How Bee compares

Bee next to Grafana Beyla, Dynatrace and Datadog.

Beyla is the open-source eBPF agent Bee's architecture follows. Dynatrace and Datadog are the platforms most teams pay for today. Bee covers fewer things than any of them, and goes deeper on the ones it covers.

Mint and a Wins badge mark the winner of the row. Tie means more than one product wins that capability.
Capability	Obstack Bee	Grafana Beyla	Dynatrace OneAgent	Datadog Agent
What you run	TieOne Rust binary per node, eBPF embedded	TieOne eBPF agent per node, open source (now OpenTelemetry eBPF Instrumentation)	OneAgent per host, reporting to a Dynatrace cluster	Agent per host, plus configuration per capability
Application code changes	TieNone	TieNone	None, but OneAgent is injected into your processes	A tracer library in each service for APM spans
Where telemetry goes	TieAny OpenTelemetry collector that accepts OTLP over HTTP	TieYour OTLP or Prometheus backend	Dynatrace SaaS or a vendor-run Managed cluster	Datadog SaaS
Air-gapped operation	Not on list plans; quoted as Enterprise (offline delivery and updates)	Possible; you assemble the stack	Managed option, still vendor-operated	No
Distributed tracing	WinsIntra-service, intra-host and cross-host, no coordinator	Propagated context; you assemble the backend	In-process agent plus a Dynatrace cluster	Tracer libraries plus Datadog APM
Profiling	WinsTied to the same request as the trace	Not a first-class profiler	Code-level through the in-process agent	Separate profiler product
Named Node.js async waits	WinsYes, tied to the exact request that caused them	No, generic transactions	Method-level through the in-process agent	Through tracer and profiler integrations
Full MySQL query text	WinsYes, with typed prepared-statement parameters	Generic SQL metadata	Through the in-process agent	Through tracer libraries
Trace correlation	WinsEvidence only; refuses to guess	Heuristics plus propagated context	In-process agent context	Tracer-propagated context
Resource ceilings	WinsPublished for every queue, map and buffer	Partial	Vary by agent configuration	Vary by agent configuration
Cost during an incident	WinsFlat per month, unchanged; the node count is not on the invoice	Free software; backend costs are yours	Consumption units rise with load	Host, span and ingest meters rise with load
Output format	TieOTLP and Prometheus	TieOTLP and Prometheus	Proprietary ingestion	Proprietary ingestion

Swipe the table sideways to reach every column.

Where the others win

We would rather you hear it from us. Pick something else if you need:

PostgreSQL or MongoDB today. They are on Bee’s roadmap; Beyla covers them now, and ships in the Obstack Stack.
ARM64, Windows, or any distro other than Ubuntu 24.04 and 26.04.
Framework-aware spans inside Java, .NET or Python services.

Based on each vendor's public documentation, reviewed September 2026.

05 / Runs on

A package per node. Pods stay untouched.

Ubuntu 24.04 LTS.deb

Ubuntu 26.04 LTS.deb

Kubernetesone agent per Ubuntu node

Requiresx86-64 · cgroup v2 · BTF

License key, then git

The first month is a key. Source follows the agreement.

The first month Bee runs under a license key, as Ubuntu packages. After that the PolyForm Internal Use License applies: we send an agreement, you sign it, and we grant git access to the source. Use and change it for your own operations; do not distribute it. Stop paying and git access stops.

$ sudo apt install ./obstack-bee_0.2.4_amd64.deb
$ curl http://127.0.0.1:9400/healthz

06 / Pricing

Flat per month. Bee on every node you run.

Bee is not licensed per agent. Every plan carries it on as many Linux hosts, virtual machines and Kubernetes workers as you like, with distributed tracing and profiling, intra-service and intra-host, all protocols and Node.js await naming. What separates the plans is the backend behind Bee and the support that answers you.

Where the backend runsSame software, same modules, same support. Only the hardware bill moves.

Pick your own hardware to take the hosting off the price: $120€100 off Go, $360€300 off Pro, $1,080€900 off Max. Starter stays Obstack Cloud only.Showing on-premise prices — the list price less the hosting we no longer buy: $120€100 off Go, $360€300 off Pro, $1,080€900 off Max. Starter is Obstack Cloud only and is unchanged.

Best effort · Business hours

Starter

$59€49/ month

1 TB storage included · no add-ons

Obstack Cloud only — no on-premise optionObstack Cloud only

The stack on shared cloud capacity, for a first fleet or a side estate that can live with best-effort ingest.

Shared infrastructureMulti-tenant capacity on Obstack Cloud

Rate limited, best effort. Excess batches are shed, never billed. Your collector retries on 429 Too Many Requests and 503 Service Unavailable.

The full stack on Obstack Cloud, on shared capacity
Unlimited nodes and unlimited Grafana users
Rate-limited, best-effort ingest: excess telemetry is shed, not billed
1 TB of object storage included
Email support, best effort, no response-time commitment
Bee, Eagle, Grafana and EagleMCP included; no paid add-ons

See Starter

Tier 1 · Business hours

Go

$599€499$479€399/ month

Flat on Obstack Cloud · hardware includedFlat on your own hardware · $120€100 hosting credit applied

A backend of your own, deployed and supported, for teams that page each other in business hours.

Dedicated node, no HAOne node · 4 cores · 32 GB RAM · 500 GB NVMe

The full stack: Bee, Beyla, Vector, GreptimeDB, Grafana, Alertmanager
Unlimited nodes, unlimited ingest, unlimited Grafana users
One dedicated backend node: 4 cores, 32 GB RAM, 500 GB NVMe
10 TB of object storage included
Setup, updates, rollout and monitoring of the infrastructure
Email support, next business day response
EagleSRE production add-on at launch; preview by request
Bee, Eagle, Grafana and EagleMCP included

See Go

Tier 2 · Business hours

Pro

$1,799€1,499$1,439€1,199/ month

Flat on Obstack Cloud · hardware includedFlat on your own hardware · $360€300 hosting credit applied

Everything in Go on a much larger node, plus engineers who answer the same working day.

Dedicated node, no HAOne node · 16 cores · 64 GB RAM · 8 TB NVMe

Everything in Go, same unlimited nodes and ingest
A larger dedicated node: 16 cores, 64 GB RAM, 8 TB NVMe
Eagle console included: workloads, logs, traces and Ask Eagle
Tier 2 support during working hours, P1 answered in 2 business hours
Quarterly stack review and upgrade planning
10 TB of object storage included
EagleSRE production entitlement included at launch; preview by request

See Pro

Tier 3 · 24/7, every day of the year

Max

$5,399€4,499$4,319€3,599/ month

Flat on Obstack Cloud · hardware includedFlat on your own hardware · $1,080€900 hosting credit applied

Everything in Pro on a high-availability cluster, with round-the-clock Tier 3 engineers and EagleSRE production entitlement at launch.

Three-node HA cluster3 nodes · 16 cores · 64 GB RAM · 8 TB NVMe each

Everything in Pro, Eagle included
Three-node HA cluster: 16 cores, 64 GB RAM, 8 TB NVMe each
EagleSRE production entitlement included at launch; preview by request
Tier 3 support 24/7, P1 answered in 30 minutes, any day of the year
Phone escalation and a 99.9% availability commitment on Obstack Cloud
10 TB of object storage included
Priority on roadmap requests

See Max

Tier 3 · Named engineer · 24/7Quoted

Enterprise

A tailor-made solution when the list tiers do not describe your environment.

Sized with you: cluster shape, regions and retention designed around your estate.

Air-gapped delivery and offline updates
Multi-region and multi-tenant deployments
Custom model hosting for Eagle and EagleSRE
Custom SLA with a named engineer
Security review pack, DPA and procurement paperwork

Customprice

Quoted against your estate

Request a quote

One price, whatever you runNo node, user, ingest, egress or query meters

FlatAny fleet size, any traffic

10% offYearly billing

15% off3-year term, price locked

30 daysMax demo license · self-hosted Helm chart

See the plans

Go, Pro and Max include 10 TB of object storage in the EU or US region you pick, and can buy more at $179€149 per 10 TB per month, egress and API calls included. Starter includes 1 TB and takes no add-ons. On your own hardware you supply the nodes and the bucket, and the hosting comes off the price.