latitude vs future agi

Two Open-Source Platforms of Different Widths

Latitude, from Barcelona, is open-source observability for agents: "Latitude traces your agent in production, finds the failures, and dispatches your coding agent to fix them." Its V2, released in June 2026, is deliberately narrow. It traces sessions, tools, memory and cost over OpenTelemetry with about 40 integrations; evaluates completed sessions with rule conditions, an LLM judge or custom JavaScript; supports human annotation; alerts through monitors; and sends Claude Code or Cursor to open a fix PR through Agent Dispatch. V2 dropped V1's prompt gateway, triggers and hosted tools; V1 keeps running for existing customers.

Future AGI is open source at the other extreme of scope. Its platform, Apache-2.0 at the core since Q2 2026, has six parts: Simulate (multi-turn text and voice), Evaluate (built-in evaluators, "50+ metrics" by its README's count, and its own Turing judge models), Monitor (traceAI on OpenTelemetry, and Error Feed), Protect (guardrails that block, warn, mask or log), the Agent Command Center gateway to 100+ model providers, and Optimize (prompt optimization). It reports "10K+ self-hosted deployments", by its own count.

If you are choosing between them, you have probably already decided you want to run the thing yourself, or at least keep the option. So this page starts with licences and servers and gets to features after.

Side by Side

Future AGI has the longer feature list by a wide margin. Latitude has the simpler licence and the more finished Kubernetes story. For a platform team, the second pair of facts often matters more than the first.

MIT, or Apache-2.0 with an EE Directory

Latitude's licence is the easy read. The latitude-llm repository moved from LGPL-3.0 to MIT on May 20, 2026, and the platform in it is the platform Latitude runs: its deployment docs say every self-host option "deploys the same images with the same configurable values". There is no enterprise directory to check. Latitude's Enterprise tier sells custom on-premises deployment, SAML SSO, RBAC and an SLA, which are services and support around the same code.

Future AGI's licence needs one more step. The core is Apache-2.0, a permissive licence with an explicit patent grant that many legal teams like. Beside it sits a LICENSE-EE file: code under the repository's ee/ directories needs an Enterprise licence key or a Future AGI Cloud plan in production. That is a standard open-core arrangement. It does mean that before your teams build on a feature, someone should check whether its code lives under ee/.

Self-Hosting One by Friday

A platform engineer is asked on Monday to get one of these running on the company's own infrastructure by Friday, for a pilot with three agent teams. Here is the week, with each vendor's documented path.

1. Monday: count the services

   Latitude runs five application services, web, api, ingest, workers and workflows, on PostgreSQL with pgvector, ClickHouse, Redis, Temporal and an object store. Future AGI's requirements page lists PostgreSQL, ClickHouse, Redis, Temporal, MinIO for object storage and a code executor that needs privileged containers, and its README also names RabbitMQ. It sizes a production install at 16+ cores, 32+ GB of RAM and 200 GB+ of SSD, and warns that "under-provisioning RAM is the most common cause of a failed first boot". Both stacks have a column store, a workflow engine and object storage. Neither is a single binary.

2. Tuesday: one host

   Both stand up on a single machine with Docker Compose. Latitude calls its Compose route "a production-grade instance on one machine". Future AGI's README starts from published images and docker compose up -d on a fresh clone, and a profile system lets you choose which services run. Set FUTURE_AGI_TELEMETRY_DISABLED=1 if you do not want the default telemetry; the README says it collects registration data and usage counts, and "No trace data, no prompts, no API keys, ever."

3. Wednesday: the production shape

   Latitude has a Helm chart for "a scalable, highly-available deployment on Kubernetes". Future AGI's docs are direct: there are no Helm charts or Kubernetes manifests yet, that support "is on the roadmap", and Docker Compose "is the only supported path today". On a Kubernetes-only platform, that is the conversation to have first, along with whether your cluster policy allows the privileged code executor.

4. Thursday: read the issue trackers

   Future AGI's open issues include a ClickHouse config crash-loop (#1540), missing gateway logs (#1060), reCAPTCHA blocking self-hosted login (#155), a bad span dropping a batch (#1080), SDK and platform evals disagreeing (#1890), and a CDN dependency breaking offline installs (#2339). Latitude's include near-duplicate failure clustering (#4396), a pinned OpenTelemetry version disabling CrewAI tracing (#4373) and token-sum mismatches (#4497). Check each against your agents' frameworks and your network rules.

5. Friday: hand it over

   On a Kubernetes shop, Latitude is the likelier Friday demo, because the chart exists. On a Compose-friendly host with enough RAM, both are realistic, and Future AGI gives the three teams far more to try.

The honest summary is that the footprints are close in kind. Future AGI's is somewhat larger, with a privileged code executor and, per its README, a message queue; but anyone who tells you one of these is light and the other heavy is comparing feature lists, not Compose files. Both are real distributed systems with a column store and a workflow engine in them, and both want the care you give any stateful service.

What You Get Once It Is Running

Here the difference is large. Future AGI gives the three pilot teams a simulation environment for testing agents before release, including voice; built-in evaluators and Turing judge models to score with; an OpenAI-compatible gateway; guardrails that can mask personal data or block a disallowed tool call in the traffic they see; and prompt optimization. Its Tool Permissions and MCP Security scanners enforce tool access for traffic routed through the gateway.

Latitude gives them less, and more finished in its lane. Evaluations come in three kinds, and its docs describe rule conditions as "deterministic checks, free and instant". Human annotation is built in, and evaluations Latitude generates stay aligned to annotations over time. Datasets built from real traces feed replays and regression tests. Monitors alert in-app, by email and in Slack. Agent Dispatch sends Cursor, Claude Code, Linear or a webhook to investigate a failure through Latitude's MCP and open a PR that you review. None of it acts on a live request: evaluations score sessions after they complete.

So the feature question comes down to one thing: do you need anything in the request path? If yes, only Future AGI has it, and you will be running its gateway. If no, Latitude's narrower scope is less to operate, less to learn, and less to upgrade.

Where Each One Is Stronger

Latitude

• Licence clarity. MIT across the whole repository, with no enterprise directory to audit.
• Kubernetes today. A Helm chart for a highly available deployment, where Future AGI supports Docker Compose only.
• The repair loop. Agent Dispatch turns a failure into a pull request someone reviews.
• Operational scope. Nothing in the request path, so nothing on the hot path to keep available.

Future AGI

• Breadth. Simulation, evaluators, a gateway, guardrails and prompt optimization in one deployment.
• Runtime control. The only one of the two that can block or mask anything in live traffic.
• Deployment choices. Managed SaaS, your own VPC on AWS, GCP or Azure, or air-gapped on-prem, besides self-hosting.
• Published compliance add-ons. SOC 2 Type II, HIPAA BAA, SAML SSO and SCIM, each with a list price.

If You Would Rather Not Run It

Both clouds start free with unlimited seats and 30-day retention, which makes a side-by-side trial cheap. Each has one caveat on its pricing page. Latitude does not say what a credit measures; the page lists the question without answering it there. Future AGI's pricing and enterprise pages disagree on some SLA figures and gating. Get both answers in writing before you model a bill.