SPECLAN v0.9.6 — Use SPECLAN With Your ChatGPT Account

If you already pay for ChatGPT Plus, Pro, or Team, you can now use SPECLAN — no separate API key, no additional subscription on top.

That one sentence is the v0.9.6 release. Everything else in this note builds out from it.

Sign In With ChatGPT — The Headline Feature

Until today, using OpenAI as your SPECLAN provider meant generating an API key in the OpenAI developer console, managing it like any other credential, and watching a per-token bill tick up on a developer account you might not even own. For the majority of ChatGPT users who never touch the developer console at all, that was a hard barrier. SPECLAN's value proposition was, effectively, hidden behind an API-key requirement that most potential users couldn't — or wouldn't — clear.

v0.9.6 removes that barrier entirely. Codex OAuth lets you sign in with the same ChatGPT account you already use in the browser or the ChatGPT app, and your existing Codex budget — the token allowance bundled with an active ChatGPT Plus, Pro, or Team subscription — powers every AI feature in SPECLAN.

The User Journey

The end-to-end experience is designed to take less than a minute:

1. Discover the option. Open the SPECLAN sidebar in VS Code, click the gear icon to enter Settings, and go to the LLM Configuration tab. The API Keys section now shows a Sign in with ChatGPT button alongside the existing API-key fields.
2. Initiate the OAuth flow. Click the button. SPECLAN opens your default browser to the ChatGPT authorization page — the same domain you already trust with your ChatGPT session.
3. Grant consent. Confirm that SPECLAN may act on your behalf for Codex API calls. If you're already signed in to ChatGPT in that browser, the consent page appears directly with no additional login step.
4. Arrive at a ready-to-use session. The browser hands you back to VS Code, the Settings panel reflects your signed-in state, and every AI feature in the extension — spec editor chat, Infer Specs, Clarification, CR Merge, HLRD Import — is immediately available. No key to paste. Nothing to copy. Nothing to rotate.

What the Codex Budget Is, and How SPECLAN Uses It

Your ChatGPT subscription includes an allocation of tokens that OpenAI bills against every time you send a message in the ChatGPT app or call their Codex API. That pool is your Codex budget. When SPECLAN routes an AI call through your OAuth session — whether you're asking the spec editor to rewrite a section or running Infer Specs across a large codebase — each call debits the same budget as a message in ChatGPT would. There is no separate billing relationship, no additional invoice, and no developer-account configuration. If you already pay for ChatGPT, your SPECLAN usage is covered by what you already pay for.

Two practical notes follow from this:

• Finance conversations are simpler. There is no new line item to justify; SPECLAN rides on an existing subscription your organization may already have approved.
• Key rotation concerns disappear. You no longer have an API key sitting in your VS Code settings — you have a session, managed by OAuth, revocable from the ChatGPT account page at any time.

Codex OAuth is the single change in v0.9.6 we expect most users to feel immediately. If you had been evaluating SPECLAN and stopped at the provider-configuration step, this is the release where you should look again.

Also In v0.9.6

Two secondary capabilities complement the headline. Both widen the set of environments where SPECLAN can run, but neither changes the default, get-started path the way Codex OAuth does.

Local LLM Support — Ollama, LM Studio, and vLLM

v0.9.6 ships the first user-visible Local LLM provider, gated behind the experimental flag. Point SPECLAN at a local server running Ollama, LM Studio, or vLLM and every AI feature in the extension routes through your own machine instead of a cloud API. For teams whose spec tree is a compliance problem if it leaves the machine, or who run Infer Specs across a large codebase often enough that cloud spend becomes a real line item, local is the answer. We've written a comprehensive Local LLM setup guide with step-by-step instructions for each supported server and honest guidance on which workloads work on which models. For the inside story of how we shipped this — the benchmark that decided our recommendations and the Mac-friendly model we ended up keeping — read the companion blog post: We Gave SPECLAN a Local Brain.

Weaker-Model Adoption — Reliability on Smaller Models

Not everyone has access to frontier models. Some teams are rate-limited, some work under an approved-model list, some are on a local LLM that has fewer parameters than a cloud giant. v0.9.6 includes a package of prompt and workflow adjustments that make SPECLAN behave reliably against weaker models, not just flagship ones: every editor transformation now carries an explicit document-completeness rule in its system prompt, structured-output schema hints are enforced with a regression test, agentic tool-call retries are bounded, and a new live-LLM test harness (apps/ai-agents-e2e) validates the prompt fixes against real local models. If you were getting mixed results on a smaller model, v0.9.6 should feel distinctly more reliable; if you were on a frontier model, you won't notice anything different.

Curious Which Model Writes the Kind of Spec Your Team Would Commit?

With ChatGPT OAuth, Local LLM support, and weaker-model reliability all landing in the same release, the question shifts from can I use SPECLAN with my provider to which provider writes the specs I actually trust. We ran the same brief — the excalidraw codebase — through Claude Opus 4.7, Claude Sonnet 4.6, Claude Haiku 4.5, GPT 5.4, and GPT 5.4 Mini and parked every resulting spec tree side-by-side at speclan.net/compare. Opus produced 197 requirements across 16 features; GPT 5.4 produced 43 across 6. Same prompt, dramatically different decomposition. Pick a node, read what each model authored, calibrate before you configure.

Get the Update

If you already have SPECLAN installed from the VS Code Marketplace, v0.9.6 will update automatically. New to SPECLAN? Install the extension, open Settings, click Sign in with ChatGPT, and you're ready.

One sign-in. Your existing subscription. Every SPECLAN feature. That's v0.9.6.