nexgenr
NexgenR CLI is a terminal AI coding agent wired to the NexgenR red-team surface.
Install
Requires Node.js 22.19.0 or later.
npm install -g nexgenr
Or with pnpm:
pnpm add -g nexgenr
Configure
NexgenR signs in with an API key. There is no OAuth and no device flow:
export NEXGENR_API_KEY="<your key>"
# optional, defaults to https://redteam.nexgen.systems/v1
export NEXGENR_BASE_URL="https://redteam.nexgen.systems/v1"
nexgenr login # reads NEXGENR_API_KEY
nexgenr login --api-key <your key> # or pass it explicitly
The same key works on both surfaces; the header is what changes:
| Surface | Send the key as | Rejected |
|---|---|---|
/v1/… inference and catalog | Authorization: Bearer <key> or x-api-key: <key> | x-bf-vk answers 401 here |
/mcp gateway | x-bf-vk: <key> or Authorization: Bearer <key> | — |
The key must be provisioned for the vllm provider and the model you use
(nexgen-redteam or refuseless-redteam). Without that grant GET /v1/models answers
200 with an empty catalog and chat fails with 400 modelo-desconocido.
nexgenr login writes ~/.nexgenr/config.toml with the redteam provider and its two
models, nexgen-redteam (262 144 tokens of context, the default) and refuseless-redteam
(1 048 576 tokens), and ~/.nexgenr/mcp.json with the gateway. It does not call the
catalog: the model list is static.
Quick start
cd your-project
nexgenr -p "Summarize this repository"
nexgenr # interactive terminal UI
What it talks to
| Capability | Endpoint |
|---|---|
| Chat, streaming only | POST /v1/chat/completions |
| Model catalog | GET /v1/models |
| Health and published limits | GET /health, no credential |
| MCP gateway | POST /mcp — 24 tools (8 Brave_Search, 16 Crawl4AI) |
POST /v1/completions and POST /v1/embeddings are in the edge allowlist but the engine
answers 404: POST /v1/chat/completions is the only inference route.
Limits, as published by GET /health: 4 requests in flight per consumer with 429 and
Retry-After: 10 beyond that, a 12 MiB request body, and 262 144 tokens of context. A
prefill can take minutes, so requests stream and wait up to 3 600 s for the first byte.
Features
- Terminal UI, headless
-pmode, ACP server, and a local REST/WS API (nexgenr web). - Sessions persisted under
~/.nexgenr/sessions, withwire.jsonlas the source of truth. - Permissions and approvals with
manual,yolo, andautomodes. - Agent Skills, MCP servers, subagents, goals, cron tasks, and Swarm fan-out.
- Provider/model configuration for any OpenAI-, Anthropic-, or Gemini-compatible endpoint.
Not included in this build
- Remote Control (requires a relay service).
- CDN-based auto-update; upgrade with
npm install -g nexgenr@latest. - The bundled browser UI;
nexgenr webserves the API plus a minimal landing page. - SSO: NexgenR signs in with an API key only.
License
MIT. See LICENSE and NOTICE for the upstream attribution.