pi-zed-provider

Use a Zed subscription as a native model provider in PI and OMP.

The extension registers a zed provider whose model catalog comes from the connected Zed account. Account storage, sign-in, token refresh, model catalog, request conversion, streaming, tool calls, and usage accounting all run inside the extension; no gateway process is involved.

Requires a Zed account with hosted-model access (Pro, Student, or Business). Not affiliated with Zed Industries.

Install

From npm, one command per host:

omp plugin install pi-zed-provider
pi install npm:pi-zed-provider

From a source checkout:

git clone https://github.com/trefeon/Zed2Api.git
cd Zed2Api
omp plugin install .
pi install .

For a single run without changing host configuration:

omp --extension .
pi --extension .

Both hosts read the entry point from package.json (pi.extensions and omp.extensions). A .tgz is not an installable source on either host; install the package or the directory instead.

Sign in

pi  # then run /login, choose Zed, and name the account
omp login zed

/zed login [name] is also available for adding or naming extra accounts. The browser completes Zed's native-app sign-in and the callback writes the account to ~/.zed2api/accounts.json with owner-only permissions.

Commands

Command	Effect
/zed or /zed status	Accounts, active account, cooldowns, last HTTP status
/zed login [name]	Add an account through the browser flow
/zed accounts	List stored accounts
/zed switch <name>	Make an account active
/zed forget <name>	Remove an account
/zed models / /zed refresh	Print or refresh the live model catalog
/zed probe [name]	Send a 32-token completion per account and report state
/zed usage [--json] [--all]	Local token spend; exact plan usage when a dashboard cookie is configured

Models

The provider advertises every model the account can serve. The request dialect is selected from the model id:

Family	Id prefix	Wire dialect	Reasoning
Anthropic	claude* and unknown ids	Messages	low … max via output_config.effort
OpenAI	gpt*	Responses	low, medium, high, xhigh
Google	gemini*	Gemini contents / systemInstruction / functionDeclarations	forwarded as supported
xAI	grok*	Chat Completions	forwarded as supported

The model list is live: a successful sign-in, /zed refresh, or a host-side refresh updates the cache in ~/.zed2api/model-catalog.json. OMP merges discovered models into its catalog; PI seeds its startup registration from the cache so later processes see the full list.

Window and output limits come from the highest specification declared for the same model in the built-in provider catalog, not from a family-wide ceiling. Current examples: GPT-5.6 and GPT-6 advertise 1,050,000 context and 128,000 output tokens; Claude Sonnet 5 and Haiku 5 advertise 1,000,000 and 128,000; Gemini advertises 1,048,576 and 65,536. New ids that are not in the static map fall back to the family's highest specification. ZED2API_CONTEXT_WINDOW and the per-family output overrides still win.

Usage

/zed usage reports the turns this provider served, using Zed's published hosted-model rates. Models that are not yet on Zed's price page use the provider list price +10%, the same markup. Long-context and date-based price tiers are not modeled; the base tier is used. A dashboard cookie adds the account-wide figures from dashboard.zed.dev, including the editor and other clients.

/zed usage
/zed usage --json
/zed usage --all

The local ledger is ~/.zed2api/usage.jsonl; no prompt or response text is stored. The dashboard is authoritative for billing. See docs/usage.md for cache accounting, plan credit, and how to supply the dashboard cookie.

Configuration

All settings are optional. Credentials live outside the repository.

Variable	Purpose
ZED_ACCOUNTS_FILE	Account-store path (default ~/.zed2api/accounts.json)
ZED_USAGE_FILE	Local usage ledger path
ZED_PLAN	pro, student, business, or free when no dashboard cookie is set
ZED_DASHBOARD_COOKIE	Cookie header for dashboard.zed.dev usage
ZED_API_BASE	Upstream base URL; used by the mock upstream in tests
ZED2API_CONTEXT_WINDOW	Override the advertised context window
ZED2API_CLAUDE_MAX_TOKENS	Override the Anthropic output cap
ZED2API_GPT_MAX_TOKENS	Override the OpenAI output cap
ZED2API_CLIENT_VERSION	Override the Zed client version sent upstream
ZED2API_HTTP1	1 forces HTTP/1.1 for the HTTPS edge (debugging only)
ZED2API_SYSTEM_ID	Override the per-install system id

Development

npm ci
npm run typecheck
bun run test
npm run pack:check

The tests use a mock upstream and stub transport; no Zed account or network is required. See docs/testing.md for the test harness.

Host-specific install and login details: docs/install-pi.md, docs/install-omp.md, and docs/host-login.md.

Security

Account credentials and usage data stay on the local machine. The provider does not log prompts, responses, or tokens. Do not paste accounts.json, dashboard cookies, or full request dumps into an issue. Report vulnerabilities through GitHub Security Advisories as described in SECURITY.md.

License

MIT. See LICENSE.