Agent Studio

Ten focused tools + four advanced multi-step agents with a real tool runtime (not one-shot email wrappers).

Advanced runtime

Agents can plan → call tools → remember → critique → finalize, and the UI shows every step.

Tool	Purpose
web_search	Live web search
fetch_url	Read a page
calculate	Safe math
memory_set / memory_get	Short-term memory
save_artifact	Persist markdown under runs/
request_critique	Second-pass critic LLM

Advanced agents

Slug	What
operator	General goal → autonomous tool loop
deep	Multi-round research + critique + revise
swarm	Planner → workers → merger
builder	Spec → researched implementation plan

Focused tools

brief pitch triage watch actions onboard debug publish study remit — single-purpose workflows (still email-capable).

Setup / run

source ~/.venvs/research-agent/bin/activate
pip install -r requirements.txt
cp .env.example .env   # GROQ_API_KEY + SMTP_*

./run-api.sh
cd web && npm run dev

Open the hub → Advanced runtime section first.

API

• GET /api/agents
• POST /api/agents/{slug}/run → includes meta.steps for advanced agents