Inline comments on every pull request. Second-pass self-verification drops made-up findings before they reach your PR. Every finding comes with a proof packet.
Genie spreads every request across the GPUs you already own and routes by persona, so the right model handles each task. Prompts and code never leave your network. Cloud overflow is opt-in — and runs the same models you run locally.
YOUR NETWORK
Every request lands on a GPU you own.
request→fleet router→least-loaded GPU for the persona
gpu-01
review · 72%
gpu-02
coder · 48%
gpu-03
chat · 91%
cloudoverflow · opt-in
Prompts and code stay inside your network. Cloud overflow is opt-in and runs the same models — never a third-party API.
An OpenAI-compatible API in front of vLLM, SGLang, and ktransformers. Send a persona; Genie picks the model, the quant, the GPU, and the engine that wins that workload. No vendor lock-in.
ONE ENDPOINT
OpenAI request in. The right engine out.
POST /v1/chat/completions→persona router
vLLMunique promptsROUTED
SGLangprefix-shared review
ktransformerslong context
Pick a persona — Genie picks the model, quant, GPU and engine. Drop-in OpenAI SDK, no vendor lock-in.
A knowledge layer that refuses to store a "fact" without a citation. Every recalled claim carries the source it came from and a confidence score, so an agent can verify before it acts.
EVERY CLAIM HAS A SOURCE
Memory that refuses to guess.
Prod DB is Neon, region us-east-1
↳ infra/runbook.mdconfidence 0.98
Auth uses Supabase session cookies
↳ PR #230 reviewconfidence 0.91
Rate limit is 60 rpm per API key
↳ docs/budgetsstale · re-ground
No citation, no ingest. Stale facts surface for re-grounding instead of being regenerated as truth.
Talk to Genie in the browser, or call the OpenAI-compatible API — same agent, same skills, same memory either way. Nothing to install to start a conversation.
WEB + API
Same agent. Two front doors.
You
What broke the deploy last night?
The 02:14 deploy failed on a Prisma migration lock. The stuck backend was terminated; the retry is green.
$ curl api.genie.tech/v1/chat/completions
-H "Authorization: Bearer sk-genie-…"
Use it in the browser or call the OpenAI-compatible API. One agent, one set of skills, one memory.
Save a named context blob — straight from your clipboard or a file — and pull it back from the CLI, MCP, or any chat. Org-shared and size-bounded, so the whole team works from the same source.
SHARED CONTEXT STORE
Save it once. Load it anywhere.
$ genie context save --from-clipboard --name api-spec
saved · ctx_8f2a · 14 KB · org-shared
$ genie context get -c api-spec
→ copied to clipboard · ready to paste into any chat
A named, size-bounded context blob — from your clipboard or a file. Shared across your org, loadable in CLI, MCP and chat.