Middleware is a full-stack observability platform founded in 2022 that covers infrastructure monitoring, APM, logs, real user monitoring, synthetic checks and LLM observability on OpenTelemetry. Its OpsAI agent detects production errors, traces the root cause and opens a pull request with a fix. The vendor lists more than 350 integrations.
About Middleware
Middleware is a full-stack observability platform from a San Francisco company founded in 2022 by engineers from Netflix, DigitalOcean, Google, Cisco and Freshworks. It collects logs, metrics, traces and real user sessions into one timeline, so a team can follow a slow checkout from a browser click down to the database query behind it. It is a Y Combinator Winter 2023 company, and it has moved from classic monitoring toward an AI agent that acts on what it finds.
The agent is called OpsAI. It watches traces from application performance monitoring, browser sessions, Kubernetes events and alerts, finds the root cause of an error and opens a pull request with a proposed fix. For Kubernetes it offers two modes: Auto RCA, which suggests a fix for a person to review, and Auto Fix, which applies the change directly. It can also ingest alerts from Datadog and Grafana and investigate them with its own data, so a team can try it without migrating anything. Teams that run it inside Slack can tag the bot on an alert and let it trace the problem.
The rest of the suite covers infrastructure monitoring, APM, log monitoring, real user monitoring, synthetic checks and browser tests. A separate LLM observability product traces prompts, token use, evaluation scores and GPU metrics for AI applications and agents, and it works with any provider that supports OpenTelemetry. That makes it a reasonable single pane for teams that run both classic services and model-backed features. If you only need LLM evaluation, a dedicated tool such as Confident AI or Openlayer goes deeper on test suites.
The pricing model is usage based: a free plan with monthly limits, pay as you go per gigabyte of logs, metrics and traces, and a custom BYOC tier. For incident response specifically, compare it with Rootly, which manages the human process around an outage rather than the telemetry itself, and with Cicube for build pipeline health.
The main caveat is support. Free users get community help only, and round-the-clock support is attached to the custom tier. Two of the vendor's own pages disagree on the per-gigabyte price, so confirm the figure on the live pricing page before you budget.
Pricing
Free Forever is $0 with a monthly allowance of 100GB data, 1K RUM sessions, 20K synthetic checks, 10 browser test runs and 2 million OpsAI tokens, kept for 14 days. Pay As You Go charges $0.3/GB for metrics, logs and traces, meters RUM, synthetic and browser usage separately and keeps data 30 days. BYOC is custom.
OpsAI detection is free; fixes are billed per token.
| Tier | Monthly price | What it includes |
|---|---|---|
| Free Forever | Free | |
| Pay As You Go | Custom | |
| BYOC | Custom |
Key Features
- OpsAI SRE agent: Detects errors from APM traces and RUM sessions, finds the root cause and drafts a code fix for review.
- Kubernetes Auto RCA and Auto Fix: Offers a review-first mode that suggests fixes and an automatic mode that applies them to the cluster.
- Third-party alert ingestion: Pulls alerts from Datadog and Grafana into OpsAI and investigates them using their metrics, logs and traces without a migration.
- LLM observability: Shows token use, cost, latency and evaluation scores across 50+ pre-built graphs, with GPU metrics correlated to traces.
- Unified telemetry: Combines logs, metrics, traces, RUM, synthetic checks and browser tests built on OpenTelemetry collectors.
- Ingestion control: A data pipeline lets teams drop data before storage, and usage is billed on what is actually stored.
Pros
- The free plan allows unlimited users and a generous data allowance, enough for a small team to run production monitoring without a card.
- OpsAI goes beyond alerting by opening pull requests, and it accepts alerts from Datadog and Grafana so teams can trial it alongside an existing stack.
- One platform covers infrastructure, APM, logs, RUM, synthetic checks and LLM observability, so teams avoid stitching together separate vendors.
Cons
- Free users get community support only, and 24x7 support is listed on the custom BYOC tier.
- Retention is 14 days on the free plan and 30 days on pay as you go, which is short for teams that need long audit trails without a contract.
- The vendor's llms.txt and its live pricing page quote different per-gigabyte prices, so budgeting needs a check against the checkout page.
Data Handling
- Data retention
- 14 days
- Compliance
- SOC 2 Type II · GDPR · HIPAA · ISO 27001 (vendor-stated in its llms.txt)
Frequently Asked Questions
What does Middleware cost beyond the free plan?
Beyond the free allowance, ingestion is metered at $0.3 for every gigabyte stored. Extra RUM sessions cost $1 per thousand, synthetic checks $1 per five thousand and browser runs $10 per thousand. Custom BYOC contracts are quoted individually.
What are the limits of the Middleware free plan?
The free plan keeps unlimited users but caps volume, with community support and two weeks of retention. Sessions, checks and browser runs also have monthly ceilings. The pricing FAQ separately mentions a trial with unlimited ingestion, so confirm which one applies at signup.
Which tools should you consider instead of Middleware?
Rootly suits teams whose pain is incident coordination rather than telemetry. Confident AI and Openlayer fit teams focused on evaluating and governing model behaviour. Datadog and Grafana remain the larger incumbents for pure monitoring breadth.
Can Middleware work alongside Datadog or Grafana?
Yes. OpsAI ingests alerts from both and runs its investigation inside Middleware using their metrics, logs and traces. That lets a team trial the agent without moving data, though full dashboards stay in the original tool until you migrate.
How do you set up Middleware's OpsAI agent?
For Kubernetes, install the Kube Agent with opsai.enabled=true. For application errors, install the APM or RUM agents and connect your code repository, which OpsAI reads through GitHub MCP. Application fixes arrive as pull requests for your team to review.
Top Alternatives
- Rootly: Pick Rootly if you need incident process and on-call workflow; pick Middleware when you also want the telemetry and a fix-proposing agent in one place.
- Confident AI: Choose Confident AI when LLM evaluation depth is the main job; choose Middleware when LLM traces must sit beside infrastructure and APM data.
- Openlayer: Pick Openlayer for ML testing and governance workflows; pick Middleware for production operations across services and models.