Sammy Coding Agent doesn't just autocomplete. It consults and implements — RAG, agents, evals, guardrails, model selection — and scaffolds production AI bots on the Sammy framework, end to end, right where you work.
Runs in
Add retrieval over my PDF manuals so users get grounded answers.
Loaded skill: rag-architecture
✓ Designed pipeline — chunking → embeddings → pgvector
✓ Wrote lib/rag/ingest.ts
✓ Wrote app/api/chat/route.ts
✓ Ran eval — answers grounded with citations
Reviewing changes…
Done. RAG is wired and passing — try a question.
What it does
Consult it on the hard AI/ML decisions, then have it implement them in your codebase.
Expert playbooks — RAG, agent design, evals, guardrails — load only when your task needs them, so every turn stays fast and focused.
It fetches live library docs to code against today's APIs and versions — past the model's knowledge cutoff, not from memory.
Chunking, embeddings, vector store, retrieval + rerank, and grounded generation with citations — designed and wired in your repo.
Tool schemas, routing, approval and safety, retries, context budgeting — the patterns that actually hold up in production.
Golden sets, LLM-as-judge rubrics, and CI regression gates so your assistant's quality is measured, not hoped for.
The moat: scaffold → wire your tools → eval a production AI bot on the Sammy framework, end to end, from inside the editor.
Every write and command is previewed and gated. Read-only steps run free; nothing changes on disk without your OK.
After real work, a fast critic re-checks the result and corrects concrete mistakes before the agent calls it done.
The moat
The agent is the front door. Behind it is the Sammy framework and cloud: model routing, billing, and a runtime your AI bots actually run on. Ask the agent to build a bot and it scaffolds the config, wires your tools, and evaluates it — on the same platform you deploy to.
// the bot-builder workflow, run for you
› sammy init — scan the codebase
› sammy generate — tools + agents
› wire to your APIs — approval-gated
› sammy eval — score & refine
✓ Production AI bot, on the platform you deploy to
Straight about scope
RAG, agents, eval harnesses, guardrails, observability, and full-stack streaming AI features — written and wired in your repo.
Fine-tuning, dataset-prep, and serving configs (LoRA/PEFT, vLLM/TGI). It authors them — it can't run GPU training in your editor.
Built on fast open models, so it's text-only today — no image input yet. It tells you when something needs your infra or a human call.
Install
Cursor, VSCodium, and Windsurf install straight from Open VSX — with auto-updates.
Cursor · VSCodium · Windsurf · Gitpod
Stock Visual Studio Code
We're finishing the Marketplace listing. In the meantime, stock VS Code users can use the manual .vsix option →
Check the listingAny VS Code-based editor
Get going in a minute
Signing in lands back in the editor you started from — Cursor, VS Code, or any fork. Free $3 credit, no card.
Add Sammy to your editor from Open VSX (or a .vsix). It lives in the sidebar.
One click opens your browser, authorizes, and drops you back in your editor — keys are minted for you, never pasted.
“Add RAG over my PDFs.” “Scaffold a support bot on Sammy.” It loads the right skill and gets to work.
Install Sammy Coding Agent and ship your first grounded AI feature today.