← Back to AI Insights
Gemini Executive Synthesis

Inability of the native `fx` CLI to use direct vendor API keys (OpenAI, Anthropic, Google Gemini), requiring routing through Vercel AI Gateway.

Technical Positioning
Flexible, direct, and secure access to leading AI models, respecting user's existing vendor relationships and data routing preferences.
SaaS Insight & Market Implications
This issue underscores a significant barrier to `fx` adoption: the mandatory routing of all major AI model interactions through Vercel AI Gateway. Users with existing direct vendor API keys for OpenAI, Anthropic, or Google Gemini are forced into an unnecessary intermediary, complicating billing, data governance, and potentially introducing latency. This design choice limits `fx`'s appeal to developers and enterprises prioritizing direct control over their AI infrastructure and data flows. The market demands flexibility; requiring a Vercel account or Gateway API key for fundamental AI access creates friction. Implementing first-class direct vendor API key backends is crucial for `fx` to become a truly agnostic and competitive AI agent platform, expanding its addressable market beyond the Vercel ecosystem.
Proprietary Technical Taxonomy
Vercel AI Gateway native CLI vendor API keys directly loopback adapter fx's Gateway protocol Vercel login AI_GATEWAY_API_KEY VERCEL_OIDC_TOKEN

Raw Developer Origin & Technical Request

Source Icon GitHub Issue Aug 19, 2026
Repo: vercel-labs/fx
Support direct vendor API-key backends

## Problem

fx can select OpenAI, Anthropic, and Google models through Vercel AI Gateway, but the native CLI cannot use those vendors' API keys directly. Users who already have vendor credentials, need direct billing/data routing, or do not want a Vercel account must build and maintain a loopback adapter that speaks fx's Gateway protocol.

The documented credential sources are Vercel login, `AI_GATEWAY_API_KEY`, and `VERCEL_OIDC_TOKEN`; there is no first-class `OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, or Gemini/Google API-key backend.

## Request

Add first-class direct vendor API-key backends, initially:

- OpenAI API
- Anthropic API
- Google Gemini API

A possible UX would be provider-specific setup such as `fx setup openai`, `fx setup anthropic`, and `fx setup google`, with secrets kept in profile-owned secure storage and the selected backend/model shown by `fx status` and `/models`.

The provider boundary should preserve the existing agent loop across backends: streaming, tool calls, reasoning controls where available, vision where available, usage reporting, ACP, and subagents. Project `.fx.json` should not be able to inject endpoints or select credential sources.

## Why this is separate from existing work

- #112 requests an OpenAI-compatible transport for **local models**.
- Draft PR #148 adds an OpenAI-compatible URL/key backend and OAuth backends for ChatGPT, Grok, and Cursor.

Those may cover direct OpenAI and OpenAI-compatible servers. This request is specifically fo...

Developer Debate & Comments

No active discussions extracted for this entry yet.

Adjacent Repository Pain Points

Other highly discussed features and pain points extracted from vercel-labs/fx.

Extracted Positioning
Skill discovery mechanism that rejects symlinks resolving outside the home or workspace root, specifically impacting Nix-managed skills.
Robust and flexible skill management, compatible with modern package managers and diverse file system structures.
Extracted Positioning
Hard-wiring of `fx` to Vercel AI Gateway, preventing host-configured custom AI model providers for `libfx` and WASM embeddings.
Open, extensible AI agent platform supporting diverse model backends and deployment scenarios (e.g., browser extensions, self-hosted LLMs).
Extracted Positioning
Lack of Homebrew Tap for simplified installation on macOS.
Standardized, user-friendly installation experience on macOS.
Extracted Positioning
Unconditional indexing of compatibility skill roots from other agent harnesses, lacking user control to disable or exclude them.
Configurable and efficient agent skill management, allowing granular control over skill discovery and loading sources.
Top Replies
boozedog • Aug 19, 2026
@ddalcu i took a stab at this on my fork, this is early version that only supports /chat/completions and not /responses https://github.com/boozedog/fx/tree/feature/openai-compatible-transport
boozedog • Aug 19, 2026
i've got a draft PR for (only!) adding openai support https://github.com/vercel-labs/fx/pull/168 this should work fine with local models too!

Frequently Asked Questions

Market intelligence mapped to Inability of the native `fx` CLI to use direct vendor API keys (OpenAI, Anthropic, Google Gemini), requiring routing through Vercel AI Gateway..

What problem does Inability of the native `fx` CLI to use direct vendor API keys (OpenAI, Anthropic, Google Gemini), requiring routing through Vercel AI Gateway. solve?
Based on our AI analysis of the original developer request, its primary technical positioning is: Flexible, direct, and secure access to leading AI models, respecting user's existing vendor relationships and data routing preferences.
What is the general sentiment around Inability of the native `fx` CLI to use direct vendor API keys (OpenAI, Anthropic, Google Gemini), requiring routing through Vercel AI Gateway.?
Yes, we have tracked 1 direct responses and active debates regarding this specific topic originating from GitHub Issue.
What architecture is tied to Inability of the native `fx` CLI to use direct vendor API keys (OpenAI, Anthropic, Google Gemini), requiring routing through Vercel AI Gateway.?
Our proprietary extraction maps Inability of the native `fx` CLI to use direct vendor API keys (OpenAI, Anthropic, Google Gemini), requiring routing through Vercel AI Gateway. to adjacent architectural concepts including Vercel AI Gateway, native CLI, vendor API keys directly, loopback adapter.

Engagement Signals

1
Replies
open
Issue Status

Cross-Market Term Frequency

Quantifies the cross-market adoption of foundational terms like streaming and vision by tracking occurrence frequency across active SaaS architectures and enterprise developer debates.