Open Source Datadog Alternatives: 9 Self-Hosted Picks

The community edition is free and runs on your own infrastructure. The Cloud Teams plan starts at $49 per month, which includes $49 of usage, billed at $0.30 per GB ingested for logs and traces and $0.10 per million metric samples, with no per-host and no per-user fees on SigNoz Cloud. A team running 20 services on 20 hosts pays the same as one running 20 services on 2 hosts, because the meter follows data volume, not infrastructure count.

Key Features

• APM covering P99 latency, Apdex, database calls and external calls per service
• Infrastructure monitoring for Kubernetes, hosts and cloud metrics alongside every service
• Alerts on metrics, logs, traces, exceptions, anomaly detection and Apdex, routed to Slack, PagerDuty, Opsgenie, MS Teams, email or webhooks
• MCP server for bringing telemetry into AI coding agents, plus a cloud-only AI teammate for incident investigation and alert tuning

Pros

• No per-host and no per-user pricing on SigNoz Cloud means the bill follows data volume rather than host or seat count
• OTel-native from the start, so there are no proprietary agent formats to migrate away from later
• Cloud is SOC2 Type II and HIPAA compliant, with data centers in the US, EU and India

Cons

• Self-hosted community edition means your team handles storage and upgrades for the backend, and that is engineering time Datadog does not ask for
• Cloud Enterprise starts at $4,000 per month, so compliance-heavy teams should factor that into the comparison before assuming the per-GB pricing is the whole story
• The MIT licence covers code outside the ee/ and cmd/enterprise/ directories, and code inside those directories is under a separate enterprise licence, so inspect those paths before assuming the entire binary is MIT

License & Hosting

• License: MIT for community code outside the ee/ and cmd/enterprise/ directories. Code in those directories is under a separate enterprise licence. Modify and self-host the community code freely.
• Where it runs: self-hosted on Docker, Kubernetes or Linux, or on SigNoz Cloud
• Setup: Docker Compose with Docker Engine 20.10+ and at least 4 GB memory. Kubernetes and Linux binary installs documented separately
• Costs: community edition costs only the VPS or cluster it runs on. Cloud Teams plan from $49/month including usage. Cloud Enterprise from $4,000/month

Best For

I'd start with SigNoz when the team needs APM, distributed tracing, log management and infrastructure monitoring in one product and wants a migration path that does not require cutting over from Datadog all at once. The Datadog Receiver keeps both platforms running during the transition. For teams whose primary need is uptime and on-call without APM, OneUptime covers that job and adds status pages and incident timelines in one product.

The honest trade-off is that this is several projects run together, not 1 binary. The Grafana OSS ecosystem covers metrics (Prometheus, Mimir), logs (Loki), traces (Tempo), profiling (Pyroscope), frontend observability (Faro), load testing (k6), and OpenTelemetry collection via Alloy. Teams leave Datadog for Prometheus and Grafana, then find themselves running a Prometheus cluster. Far cheaper, and far more work.

Key Features

• Prometheus: multi-dimensional data model, PromQL query language, HTTP pull model for metrics collection, designed for reliability as the system to consult during an outage
• Loki: horizontally scalable log aggregation, indexes only labels rather than full log text, 100% persistence to object storage for petabyte-scale storage
• Tempo: distributed tracing backend requiring only object storage, ingests Jaeger, Zipkin and OpenTelemetry trace formats
• Mimir: scales to 1 billion active series with high availability and multi-tenancy
• Grafana Alloy: open source OpenTelemetry collector with built-in Prometheus pipelines
• Grafana Cloud free tier: 10,000 Prometheus metrics, 50 GB logs, 50 GB traces and 500 VUh k6 testing at no cost

Pros

• No per-host fees on any component: Prometheus, Loki, Tempo and the Grafana OSS stack are free open source projects
• Grafana Cloud's free tier is a real starting point, not a trial limit

Cons

• Grafana is AGPL-3.0. Running it for your own team asks nothing of you. If you change the code and offer it to others over a network, you share your changes.
• You run Prometheus, Loki, Tempo and Grafana as separate projects, each with its own storage, scaling and upgrade path, and that operational cost is real and grows with usage
• Prometheus is not suitable when 100% accuracy is needed, such as for per-request billing, because the collected data may not be detailed and complete enough
• Grafana requires a database: MySQL 8.0+ or PostgreSQL 12+ for production. SQLite is not recommended for production.

License & Hosting

• License: Grafana is AGPL-3.0. Prometheus is Apache-2.0. Apache-2.0 is permissive.
• Where it runs: self-hosted on Debian, Ubuntu, RHEL, Fedora, SUSE, macOS or Windows, or via Grafana Cloud (hosted managed option)
• Setup: Grafana minimum is 512 MB RAM and 1 CPU core. Requires a production database (MySQL 8.0+ or PostgreSQL 12+).
• Costs: OSS components cost only their infrastructure. Grafana Cloud has a free tier and paid plans above it.

Best For

The Grafana and Prometheus stack suits teams already running Prometheus on Kubernetes who want to extend to logs and traces without learning new query languages for metrics. Loki uses LogQL for log queries, which is its own language to pick up. For teams who want 1 binary to operate rather than multiple services, OpenObserve covers that job.

The storage design is the other sharp difference. OpenObserve says Parquet columnar storage on an S3-native backend cuts storage costs by up to 140x against Elasticsearch, and it runs in production in environments processing in excess of 5 PB/day. All ingested data is immutable: once it is in, only entire retention periods can be dropped, not individual records. Teams with a compliance requirement for immutable audit data get that by default. Teams who need to delete a misconfigured ingestion run do not get what they want.

Key Features

• Logs, metrics, traces, RUM, dashboards, alerts, pipelines and AI observability in one binary
• SQL for logs and traces, SQL or PromQL for metrics, no proprietary query language
• Parquet columnar storage on an S3 backend
• OpenTelemetry native with no vendor lock-in
• Auto-correlation engine linking logs, metrics, traces, RUM, session replay and LLM calls into a single session
• AI SRE agent for incident triage, root cause correlation and postmortem drafting

Pros

• No per-host, per-seat or per-user fees: pricing charges only for what is ingested and queried
• Single binary with Docker one-liner install keeps operational overhead low
• SOC 2 Type II and ISO 27001 certified, GDPR compliant, HIPAA ready with BAA on Enterprise contracts

Cons

• Enterprise Edition is under a commercial Enterprise License Agreement, not AGPL: teams that need enterprise SSO (OIDC, OAuth, SAML 2.0, LDAP/AD) or federated search must move to that licence
• Cloud Professional charges $0.50 per GB ingested with annual commitment, plus $0.01 per GB queried, and at high query rates the query fee adds up

License & Hosting

• License: AGPL-3.0 for the open source edition, the same terms as Grafana (see How to choose). The Enterprise Edition is under a separate commercial Enterprise License Agreement, not AGPL.
• Where it runs: self-hosted as a single binary or Docker container, or on OpenObserve Cloud
• Setup: single Docker command using the public ECR image, no complex cluster setup required
• Costs: self-hosted open source edition costs only its infrastructure. The free tier includes up to 50 GB/day of ingestion. Cloud Professional at $0.50/GB ingested with annual commitment, plus $0.01/GB queried

Best For

OpenObserve fits when you want one binary covering logs, metrics, traces and RUM with per-GB pricing and no per-host fees. If frontend observability matters alongside backend telemetry, the built-in RUM and session replay mean you are not stitching in a separate tool. Fine-grained data deletion is not possible once data is ingested, and if your workflows depend on it, ClickStack covers logs and traces well and adds Lucene search syntax if that is the query pattern you prefer.

The part worth noting is the MCP server. Compatible agents can query telemetry, discover sources, and manage saved searches, dashboards and alerts over MCP. ClickHouse's own benchmark reports that agents using ClickStack's purpose-built MCP tools were 17% more accurate, made 25% fewer tool calls and were 2.7x more consistent than agents using a generic ClickHouse MCP server.

Key Features

• Logs, metrics, traces and session replays on ClickHouse
• Full-text search with Lucene-style syntax or SQL, with inverted indexes on log bodies and attributes and Bloom filter indexes on high-cardinality columns such as trace IDs
• MCP server for AI agent access to telemetry, saved searches, dashboards and alerts
• Schema-flexible UI: supports existing ClickHouse telemetry tables with custom field mappings, not only the OTel schema
• All-in-one Docker image bundling ClickHouse, the UI, the OTel Collector and MongoDB for local evaluation

Pros

• MIT licence on the UI and Apache-2.0 on ClickHouse and the OTel Collector: permissive for commercial use with no source-disclosure requirement
• Lucene-style log search is familiar to teams coming from Kibana or Elasticsearch
• ClickHouse reports 10 to 30x OpenTelemetry compression rates for ingestion and storage
• Hosted HyperDX Starter plan at $20/month covers 50 GB with unlimited users, a concrete comparison point against Datadog's per-host meter

Cons

• The standard open source deployment uses MongoDB to store dashboards, alerts and user profiles alongside ClickHouse: 2 database systems to operate, not 1
• Self-hosting ClickStack is MIT and Apache 2.0 software. You pay for the machines, and your team owns capacity planning, security, backups, upgrades and performance tuning for all components
• ClickStack collects anonymized usage data by default. Opt out by setting USAGE_STATS_ENABLED=false

License & Hosting

• License: UI is MIT. ClickHouse and the OpenTelemetry Collector are Apache-2.0. Both are permissive for commercial self-hosting.
• Where it runs: self-hosted via Docker, or hosted on HyperDX cloud (hyperdx.io)
• Setup: Docker with 4 ports available: 8080 (UI), 4317 (OTLP gRPC), 4318 (OTLP HTTP), 8123 (ClickHouse HTTP)
• Costs: self-hosted ClickStack is MIT and Apache 2.0 software. You pay for the machines. HyperDX hosted Starter at $20/month for 50 GB, $0.40/GB above that. Enterprise at custom pricing

Best For

Start with ClickStack when the team already runs ClickHouse or needs log and trace search with Lucene syntax and AI agent tooling over MCP. The schema-flexible UI is also useful when you have existing ClickHouse telemetry tables in custom schemas. Skip it if the team needs a single binary with no secondary databases: OpenObserve covers that job.

Check your kernel version before you deploy. Coroot requires Linux 4.16 or later, and Docker-in-Docker environments like MiniKube aren't supported.

Key Features

• eBPF-based automatic capture of metrics, logs, traces and profiles with no application code changes
• Continuous profiling down to the precise line of code for CPU and memory spikes
• SLO tracking, smart alerting and AI-powered root cause analysis
• Deployment tracking and cost monitoring
• Kubernetes Operator, OpenShift, Docker, Docker Swarm, Amazon ECS on EC2, Ubuntu/Debian, RHEL/CentOS and Windows install targets, with EKS (including basic AWS Fargate support), GKE, AKS, OKE, K3s and MicroK8s all supported

Pros

• Zero instrumentation changes means a team can get service maps and dependency graphs without touching application code
• Apache-2.0 community edition is free and covers core features with no licence fee
• Self-hosted, so data stays within your own infrastructure
• 14-day free trial with no credit card and full feature access

Cons

• Linux kernel 4.16 or later is required, so teams on older kernels are out
• eBPF does not work in Docker-in-Docker (MiniKube) or WSL1, so your local dev environment may not match production
• Standard plan costs $1 per monitored CPU core per month. A 20-core cluster runs $20/month, predictable but not free once the fleet grows
• Enterprise Premium pricing requires contacting sales

License & Hosting

• License: Apache-2.0 for the community edition. Permissive for commercial self-hosting with no source-disclosure requirement.
• Where it runs: self-hosted on Kubernetes, Docker, Amazon ECS, bare metal Linux or Windows
• Setup: Kubernetes, Kubernetes Operator, OpenShift, Docker, Docker Swarm or Linux package. Requires Linux kernel 4.16+
• Costs: community edition costs only infrastructure. Standard plan at $1 per monitored CPU core/month. Enterprise Premium at custom pricing

Best For

Coroot makes sense when the team runs Kubernetes and wants automatic service maps, SLOs and continuous profiling without touching application code. eBPF handles the observation at the kernel level, so no instrumentation changes are needed in the services themselves. Verify your kernel version before committing, and confirm that managed Kubernetes clusters you use meet the requirement. For non-Kubernetes mixed fleets, Zabbix covers more ground.

Uptrace requires ClickHouse (version 26.3 or later) for telemetry storage, PostgreSQL for metadata such as metric names and alerts, and Redis for caching. For smaller teams, that dependency chain is the real setup cost to weigh.

Key Features

• OpenTelemetry APM covering traces, logs and metrics ingested from OpenTelemetry, Prometheus, Vector, FluentBit, CloudWatch and more
• 12 supported languages: Go, Python, Ruby, Node.js, .NET, Java, Erlang, Elixir, Rust, PHP, C++ and Swift
• SAML and OIDC single sign-on documented for Google, Okta and Keycloak
• Alerting with notifications by email, Slack, webhook and AlertManager
• Flat self-hosted edition pricing: Community at $0 forever, Starter at $39/month (up to 50 GB/day), Team at $199/month (up to 200 GB/day), Business at $499/month (up to 500 GB/day)

Pros

• Self-hosted editions use flat pricing per daily volume, so the bill does not grow with every gigabyte ingested
• Uptrace Cloud has no host or seat limits, so services, users and hosts can be added freely

Cons

• AGPL-3.0 licence, the same terms as Grafana (see How to choose). Check with legal for managed-service use cases.
• Requires ClickHouse 26.3+, PostgreSQL and Redis: 3 stores to provision, back up and upgrade. Not a single-binary install
• Community edition retains data for only 14 days. Starter and above extend retention, and $39/month is the floor for teams that need longer history.
• Uptrace Cloud data is stored in EU data centers (Germany and Finland) only. Teams with US or APAC data-residency requirements must self-host.

License & Hosting

• License: AGPL-3.0.
• Where it runs: self-hosted via Docker, Ansible playbook or Kubernetes Helm chart, or on Uptrace Cloud (EU data centers)
• Setup: Docker for quick start (ClickHouse + PostgreSQL pre-configured), Ansible for bare metal, Helm chart for Kubernetes
• Costs: Community edition costs only infrastructure. Starter $39/month self-hosted.

Best For

I'd pick Uptrace when the team wants OpenTelemetry APM at a flat monthly rate on self-hosted rather than a per-GB meter that climbs with volume. The flat model pays off fast at high ingestion: on the Team tier at $199/month, 200 GB/day costs the same as 10 GB/day. Teams not already comfortable operating ClickHouse will find OpenObserve's single binary a better starting point.

Key Features

• Auto-instrument agents for Java, C#, Node.js, Go, PHP and Python. Manual SDKs for C++, Rust and Nginx LUA
• Service mesh receiver for Istio/Envoy telemetry
• AI assistant (Horizon) for querying system health with charts, topology, traces and logs
• MCP server for investigation tools and observability data from a desktop or terminal
• Multiple storage backends: ElasticSearch, H2, MySQL, TiDB, BanyanDB
• eBPF stack for Kubernetes monitoring

Pros

• Apache-2.0 licence is permissive for commercial use, modification and distribution with no source-disclosure requirement
• Apache Software Foundation project governance means stable open-source stewardship
• OpenTelemetry, Prometheus, Zipkin, Telegraf and Zabbix integrations for existing telemetry pipelines

Cons

• Storage backend selection adds complexity: the logically split architecture (Probes, Platform backend, Storage, UI) means more moving parts than a single-binary tool

License & Hosting

• License: Apache-2.0. Permissive for commercial use with no source-disclosure requirement.
• Where it runs: self-hosted. Docker one-liner script starts ElasticSearch or BanyanDB, the OAP server and Booster UI
• Setup: Docker with one-liner script for ElasticSearch or BanyanDB as storage. Multiple components (OAP server, storage, UI) to manage
• Costs: software is free. Infrastructure for storage (ElasticSearch or BanyanDB) and OAP server is the ongoing cost

Best For

Apache SkyWalking is the call when the team needs APM with broad language support and Apache Software Foundation governance, or when AI agent session monitoring is a first-class requirement: the AI Sessionizer tracks what an agent did, which tools it called and whether it met its goals. Teams that want a single binary or a hosted option will find OpenObserve covers both, and SigNoz Cloud covers the hosted option.

The community edition includes SAML and OIDC single sign-on and costs nothing beyond your own infrastructure. On-call rotation and alerts sit behind the Growth plan, which starts at $20 per user per month billed annually.

Key Features

• Monitoring for websites, APIs, ports, DNS and certificates from probes worldwide or from inside a private network, plus servers, Kubernetes, Docker, VMware, databases and network devices
• On-call paging by phone, SMS, email, push, Slack or Teams, with escalation until someone responds
• OpenTelemetry telemetry: logs, metrics, traces, exceptions, profiles and RUM in one place
• AI incident investigation with read-only root cause analysis and optional pull request creation for fixes
• MCP server for Claude, Cursor, VS Code with GitHub Copilot or any MCP client
• Status pages with 1 page and 100 subscribers on Free. Unlimited on Growth and above

Pros

• Apache-2.0 community code (outside the ee/ directory), permissive for self-hosting with no source-disclosure requirement
• One product that covers monitoring, on-call, incident management and log and trace collection, so there is one less integration surface to maintain
• Air-gapped deployment is supported with signed images, offline enterprise licence activation and no CDN dependencies, which matters for teams in regulated or network-restricted environments
• Telemetry ingest priced at $0.10/GB across all plans

Cons

• On-call rotation and alerts are behind the Growth plan ($20/user/month annual), so teams that need only on-call pay per user, which scales differently from per-GB tools
• Enterprise Edition (SCIM, audit logs, compliance dashboards) requires a licence and contact-sales pricing
• SMS alerts cost $0.10 per SMS and phone call alerts cost $0.10 per minute, and these add up in high-volume on-call environments
• The Docker Compose install asks for a Linux server with at least 8 GB of memory. The Kubernetes Helm chart route is also available

License & Hosting

• License: Apache-2.0 for all code outside the ee/ directory. The ee/ directory is under the OneUptime Enterprise License. The Community Edition image contains no code from ee/.
• Where it runs: self-hosted via Docker Compose (one Linux server, 8 GB memory minimum) or Kubernetes Helm chart. Air-gapped deployment supported
• Costs: community edition costs only infrastructure. Growth plan from $20/user/month (annual). Active monitoring from $1/monitor/month. Telemetry ingest at $0.10/GB. SMS/call alerts at $0.10 per SMS or per minute

Best For

I'd reach for OneUptime when the team needs uptime monitoring, on-call and incident management alongside logs and traces, all in one self-hosted product, especially in air-gapped or compliance-restricted environments. Running a status page tool and an observability backend as separate products means separate integration surfaces to maintain and separate bills to reconcile. Teams whose primary need is APM with deep trace analysis will be better served by SigNoz.

Zabbix collects metrics, logs, traces, and events from a long source list: OpenTelemetry, network devices, cloud services, containers, virtual machines, OS-level monitoring, log files, databases, IoT sensors, and HTTP/HTTPS endpoints. Multi-cloud cost dashboards for AWS, Azure, Google Cloud, and Oracle Cloud ship out of the box. Monitoring templates cover OpenAI Platform, Claude API, OpenRouter, GitHub Copilot, and Microsoft 365 Copilot.

Key Features

• Metrics, logs, traces, and events in one platform with event correlation, automated alerting, and remediation
• OpenTelemetry APM support for following requests across services
• Unlimited proxies for monitoring thousands of remote locations, branches, and data centers
• Business service SLA monitoring with service trees, outage simulation, and multiple status calculation algorithms
• MCP server for AI-assisted investigation and monitoring workflow automation
• Free mobile app for iOS and Android with push notifications

Pros

• The monitoring software is free under AGPL-3.0 with no paid edition: the only mandatory ongoing cost is infrastructure
• Scalable via unlimited proxy deployments across any number of remote locations
• Multi-tenancy with flexible role and permission isolation per customer, which matters for MSPs

Cons

• AGPL-3.0 applies from version 7.0: teams building a managed network monitoring service on Zabbix must make modified source available, and network-service providers should review this before deploying
• The Zabbix server requires UNIX and does not run as a server on Windows
• The UI and configuration model have a steeper learning curve than the more modern tools on this list
• OpenTelemetry is supported for APM, but tracing is not the primary focus of the product

License & Hosting

• License: AGPL-3.0 from version 7.0, GPLv2 for versions up to 6.4. The AGPL terms are the same as Grafana's (see How to choose).
• Where it runs: self-hosted on-premises or in the cloud using official packages, Docker containers, cloud images, or a pre-configured VM appliance. Zabbix Cloud is available with a 5-day free trial.
• Setup: UNIX server required. Docker, cloud image, and VM appliance installs are available. The Zabbix server does not run on Windows.
• Costs: the monitoring software is free. Zabbix sells technical support and Zabbix Cloud. Infrastructure is the only mandatory ongoing cost

Best For

I'd pick Zabbix when the primary need is infrastructure and network device monitoring across a large mixed fleet, with multi-cloud cost dashboards and no licence cost for the monitoring software. The unlimited proxy model suits distributed or multi-site environments well. For deep APM trace analysis, SigNoz is the better tool for that job.