Latitude
Open-source observability for production AI agents, combining traces, behavior search, failure signals, annotations, alerts, and coding-agent dispatch.
About Latitude
Latitude is an open-source, MIT-licensed observability platform for AI agents running in production. It records traces and spans for model calls, tools, retrieval steps, sessions, users, environments, and releases, then exposes latency, token, cost, error, and execution-path data for investigation. Teams can search real conversations, turn recurring failures into tracked signals, send alerts through Slack, email, or webhooks, and add feedback at the conversation, message, or text-range level. Reviewed examples can feed evaluation work. Latitude also dispatches Claude Code or Cursor with failure context for engineering follow-up. Starter is free and includes 20,000 credits a month with 30-day retention. Pro costs $99 a month for 100,000 credits and 90-day retention. Enterprise supports custom retention and deployment, including on-premises setups. Credit demand needs forecasting because scans and evaluation generation consume much more than ingestion. Hosted traces may contain sensitive conversations, so teams also need firm data-handling and access rules.
Best for
Best use cases
Choose it when production traces and repeated failure discovery need to share one workflow.
Consider self-hosting or Enterprise when telemetry residency and deployment control are strict requirements.
Estimate scans and evaluation generations separately from trace ingestion before selecting a hosted plan.
Compare before choosing
Unlike log-only monitoring, Latitude connects traces to behavior search, signals, annotations, and evaluation work.
Its MIT license and on-premises Enterprise option provide more deployment choice than hosted-only services.
Its credit system charges different amounts for ingestion, scanning, and evaluation generation, so headline credit totals need workload context.
Features
Investigate production agent failures
Inspect full sessions and nested spans to find failed model calls, tool errors, latency spikes, and execution paths that drive up cost.
Track recurring failures as signals
Search production traffic, group repeated behaviors or failures, and alert the team when a tracked issue appears or becomes more severe.
Review and improve agent quality
Annotate conversations or selected text, connect the feedback to signals, and use the reviewed examples to tune evaluations and confirm fixes.
Pros
- The free Starter plan includes 20,000 credits each month, unlimited seats, and 30-day retention, which lowers the cost of a team pilot.
- Open-source MIT licensing and an Enterprise on-premises option give teams more deployment flexibility than a hosted-only monitor.
- Tracing, search, signals, annotations, alerts, and coding-agent dispatch keep investigation and follow-up work in one connected workflow.
Cons
- Credits are not uniform: ingesting a trace costs 1 credit, scanning it costs 30, and an evaluation generation round costs 1,000, so workloads can be hard to forecast.
- The $99 monthly Pro plan includes 100,000 credits, but extra usage is listed at $20 per 10,000, which can raise costs for high-volume agents.
- Hosted monitoring processes traces, logs, and telemetry that may contain user conversations, so teams need data minimization and access controls before sending production traffic.
Pricing Plans
Starter
- 20,000 credits per month
- 30-day data retention
- Unlimited seats
- General channel Slack support
Pro
- 100,000 credits per month
- 90-day data retention
- Unlimited seats
- Priority support
- Extra credits at $20 per 10,000
- SOC 2 and ISO 27001 reports
Enterprise
- Custom credit volume
- Custom data retention
- On-premises or custom cloud deployment
- Fine-grained roles (RBAC)
- Team training and SAML SSO
- SOC 2 and ISO 27001 reports
- Uptime and support SLA
- Dedicated support
* Pricing information is for reference only. Please refer to the latest announcement on Latitude's official website.