AI Potluck
Product / UX / Telemetry & observability

Latitude

Latitude

Self-healing AI agent platform providing telemetry-based tracing of multi-turn sessions, tool calls and execution paths, automatic clustering of failing traces into tracked signals, coding-agent dispatch that writes and opens a fix from sample failures, and regression replay against real failing traces to verify the fix held. Self-hostable at any scale on Docker Compose, a Kubernetes Helm chart, or Railway, alongside a hosted cloud with usage-based billing.

The whole platform - web app, worker, CLI, SDKs and telemetry packages - ships MIT-licensed in one repository, with no enterprise-only directory found in the recursive tree; the hosted cloud's credit-based billing is a separate commercial layer. Only the repository is declared as an artifact. The latitude-telemetry PyPI package and the @latitude-data/telemetry npm package are one component of the platform - the instrumentation library an application imports - so their install counts measure how many applications emit traces, not how many Latitude instances run, and a component's count must not stand in for the whole platform's. The platform itself installs from Docker Hub images, a Helm chart or Railway, none of which is a declarable artifact kind. Verified 2026-09-02 via the latitude-dev/latitude-llm repository.

Openness

5 high confidence
5.0
license
MIT(OSI, single root LICENSE
source
public(apps/web, the worker, the CLI and every SDK ship in this repository
core-gated
ungated(no ee/ or enterprise-only directory found in the recursive tree

The whole platform - the web app, worker, CLI and SDKs - is MIT-licensed and self-hostable at any scale on infrastructure the README calls "fully open infrastructure": single-host Docker Compose, a Kubernetes Helm chart for a cluster, or one-click Railway. A recursive tree search for enterprise/ee/premium paths found none. The hosted Latitude Cloud sells usage credits and retention beyond the free tier, which is a separate commercial offering rather than a gate on the published source.

Adoption

2 low confidence
2.0

4,613 GitHub stars on latitude-dev/latitude-llm, which lands in the 1K-10K stars band of the stars scale, level 2. The platform is deployed as Docker Hub images, a Helm chart or a Railway template, and none of those is a declarable artifact kind, so no download count of the platform itself exists to band on. The telemetry packages (latitude-telemetry at 15,728 PyPI downloads a month, @latitude-data/telemetry at 27,563 npm downloads a month) are one component of the platform - the instrumentation library an application imports - and count instrumented applications rather than running Latitude instances, so they are recorded as context and not banded. Docker Hub reports between 22,922 and 29,501 cumulative pulls per image across the six platform images, a lifetime total that corroborates a small deployed base without being a monthly figure.

Capability

4 medium confidence
4.0

Covers tracing, dataset-based regression testing and monitoring - the dimensions this category grades - and adds an autonomous triage-and-fix loop (failure clustering, agent dispatch, verified replay) that no other product in this category offers. That is a genuinely distinctive capability set, held at level with the Langfuse anchor at 4 rather than above it: the category's top rung is reserved for Arize AX's frontier-scale closed platform, and Latitude's own dispatch loop is not yet evidenced at that scale.

  • https://raw.githubusercontent.com/latitude-dev/latitude-llm/development/README.md recorded 2026-09-02

    The Why Latitude section: "Observe: one line of telemetry captures every trace"; "Understand: failing traces are auto-grouped into tracked signals with status, size, and trend"; "Fix: Latitude dispatches your coding agents (Claude Code, Cursor) with the full context, sample traces, and a deep link. It writes the smallest correct fix and opens a PR"; "Verify: fixes are replayed against the real failing traces."

Verified 2026-09-02