grokers v0.1.1 — fail fast, fork policy, telemetry off

Changes since v0.1.0 (all verified on real runs)

• Unreachable local endpoint now fails in ~6 seconds with an actionable hint ("start ollama serve or re-run grokers-setup.sh") instead of silently retrying for ~6 minutes — the worst first-run footgun for users without ollama running.
• CONTRIBUTING/SECURITY rewritten for the fork: PRs welcome here (upstream accepts none); vulnerability reporting routed to this repo's advisories.
• All profiles now ship [features] telemetry = false in addition to auto_update = false.
• README: honest notes on residual upstream identifiers and Linux support status.

Tests: sampler/models/telemetry suites green (330 tests). Full verification battery: no-account boot, local-model coding loop, Anthropic x-api-key on the wire, dead-endpoint fast-fail.

grokers v0.1.0 — BYOK out of the box

grokers v0.1.0

The grok-build harness, unshackled: bring your own models — Anthropic, OpenAI, or a local LLM — and pay for exactly the tokens you use.

Verified in this release (real runs, not promises)

• Boots with zero xAI credentials: empty config → local ollama model completes turns (single-turn -p and full TUI).
• Real coding loop on a local model: file create + read-back + shell verify with qwen3.5:9b end-to-end.
• Anthropic Messages x-api-key fix — upstream sends Bearer for every backend; Claude BYOK would 401 without this.
• Baked BYOK model catalog (local qwen default; Claude/GPT in the picker), auto_update = false in all profiles so the upstream updater can't replace your binary.

Install

curl -fsSL https://raw.githubusercontent.com/codingmydna/grokers/main/install.sh | bash

macOS arm64 asset attached; other platforms build from source.

Fork of xai-org/grok-build @ b189869 (Apache-2.0). Independent community project, not affiliated with xAI.