Latitude V1 gateway removed: what now

the short answer

Latitude V2 removed the V1 AI gateway, PromptL, hosted tools, triggers and server-side conversation storage. Existing V1 accounts still run, but new development is focused on V2 and no shutdown date is published. Latitude recommends moving prompts into your code, calling model providers directly and wrapping those calls with capture(). Use a separate gateway if you still need centralized keys, routing, spend controls or request-time guardrails.

V2 shipped: June 2026, agent observability only
Removed: Gateway, PromptL and prompt manager, hosted tools, triggers, server-side conversations
V1 status: Keeps running for existing customers; new features ship only to V2
End date: No shutdown date is published

What V2 Removed, and What Replaces Each Part

Latitude (latitude.so) rebuilt itself as observability for agents, and its migration guide is direct about the parts of V1 that are gone. For the gateway it says: "Nothing on the Latitude side replaces them." Every row below comes from that guide.

V1 feature	V2 replacement, per Latitude
AI gateway: `latitude.prompts.run()`, `latitude.prompts.chat()`	Call OpenAI, Anthropic, Bedrock and others directly with their own SDKs
PromptL and the prompt manager	Your codebase and your version control
Hosted tools such as `latitude/extract`	A real implementation and a hand-written tool-call loop
Triggers and webhooks	Your scheduler or queue, plus Monitors and Agent Dispatch
Server-side conversation storage	Your application keeps its own message arrays
V1 evaluations	Signals and annotations

From Latitude's V1-to-V2 migration guide.

V2 is a different product direction. It keeps agent observability and evaluation, while application code now owns prompts, provider calls, tools, schedules and conversation state.

What Keeps Running

Latitude says V1 will keep running for existing customers while new features ship to V2. The migration guide does not publish a shutdown date. Before relying on V1 for another planning cycle, confirm whether it still receives security fixes, how much shutdown notice customers will receive and whether exports will remain available throughout that period.

Migrating the Way Latitude Recommends

Latitude describes V2 onboarding as a fresh start rather than a direct data migration. The practical sequence is:

Export V1 prompts and datasets as CSV.
Move prompt logic into your codebase as templates, under version control.
Call model providers directly with their native SDKs.
Instrument those calls with the Latitude telemetry SDK, wrapping them in capture() so traces carry the metadata V2 needs.
Rebuild evaluations as Signals and annotations in V2.

Exporting prompts is the easy part. Hosted tools become application code and need a tool-call loop, tests and deployment ownership. Conversation history also moves into your application, which can affect storage, retention and privacy requirements.

Keeping a Control Point in the Request Path

A gateway was never only a convenience. It was the one place every model call passed through, which made it the natural home for keys, routing, spend limits and checks. Calling providers directly removes that place. Whether you need it back depends on how many services call models and how much you want to enforce centrally.

No gateway. Direct SDK calls wrapped in capture(). Simplest, fewest moving parts, and fine for one service calling one or two providers.
A self-hosted gateway. LiteLLM exposes 100+ providers - OpenAI, Anthropic, Gemini, Bedrock, Azure and more - through the OpenAI format, with virtual keys, spend tracking, load balancing and guardrails. You keep Latitude V2 for observability and put LiteLLM in front of the models.
A gateway with guardrails attached. Future AGI's Agent Command Center is an OpenAI-compatible gateway for 100+ providers; its Protect scanners can block, warn, mask or log on model I/O, and its Tool Permissions scanner blocks tool calls for traffic routed through it.

A model gateway sees requests and responses. It may not see what an agent does next with a shell, filesystem or deployment tool. Teams with action-level risk need controls inside the agent runtime as well. Failproof AI can analyze recurring failures across agent sessions and turn a tested fix into a hook-level policy for supported harnesses, independent of which model gateway is used.

If You Are Leaving Instead

If what you valued in V1 was managed prompts, the closest open-source match is Langfuse, which has versioned prompt management with caching alongside tracing and evaluations, under an MIT core. If it was the gateway, the options above replace it with or without Latitude. If V2's observability is what you want now, staying is the least work, since your account and team carry over. The full shortlist is on Latitude alternatives.

When You Can Wait

If your V1 integration runs, you need none of V2's features, and nothing about it is on a critical path, you can leave it alone for now: V1 keeps running and no end date is published. Use the time to export prompts and datasets, and write down which hosted tools and triggers you depend on. When a date is announced, the migration is then a scheduled job rather than a fire.

FAQ

Is Latitude V1 being shut down?

No shutdown date is published. Latitude says V1 keeps running for existing customers while new features ship to V2. Confirm security-fix coverage, export access and the notice period for any future shutdown before planning around V1.

Does Latitude V2 have a prompt manager?

No. PromptL and the prompt manager are V1 features. In V2, Latitude says prompts live in your codebase and version control, and your application calls model providers directly. If you want managed, versioned prompts in a separate tool, Langfuse offers prompt management with an open-source MIT core.

Can I keep using a gateway with Latitude V2?

Yes. V2 observes calls you make yourself, wrapped in its telemetry SDK's capture(), so a gateway such as LiteLLM or Future AGI's Agent Command Center can sit between your code and the providers while Latitude traces the sessions. Latitude itself no longer provides the gateway.

Get Started

Failproof AI is free to start. It finds recurring failure modes across agent sessions using code-based and LLM-based evaluations, groups the evidence into findings, and recommends fixes. Bring the eval suite you already have, alert the right owner when behavior drifts, and turn a tested fix into a policy that prevents the failure from recurring. See pricing for the tiers.

start free get a demo →

Sources

Checked against each vendor's own site and docs on 2026-09-14. Products change; if a detail here is out of date, tell us at support@befailproof.ai.