Skip to content

🎨 Studio

Offline cockpit: kanban, live feed, markdown memory, settings. No CDN at runtime — React/Babel are vendored. Cafe Wi‑Fi can implode; Studio will not. ✈️

🕹️

Mission control vibes: start a run, watch agents stream, drag cards mid-flight, edit CONTEXT while the loop is still thinking. Feel powerful. Use responsibly. 

slmcode studio
# → http://127.0.0.1:7420
slmcode studio --listen 127.0.0.1:7421

✈️ Airplane mode

Cafe Wi‑Fi can implode. Studio will not. Bring snacks either way.

Layout 🗺️

┌──────────────────────────────────────────────────────────┐
│  Brand · Query · Run/Stop · Model                        │
├──────────────────────────────────────────────────────────┤
│  Pipeline: init → skills → … → execute → learn           │
├──────────┬────────────────────────────┬──────────────────┤
│  Nav     │  Kanban / Live / Pipeline  │  Docs / Settings │
│  Agents  │  Task inspector            │  CONTEXT/MEMORY  │
│  Skills  │  Phase · slots editor      │                  │
└──────────┴────────────────────────────┴──────────────────┘
Zone Job
🎯 Query bar Start / stop
🏭 Pipeline strip Dynamic phases from .slmcode/pipeline.yaml
🧩 Pipeline tab Bind agents per phase, insert slots, loop roles
📡 Live @agent, scope, patches, slots, output
📋 Kanban Drag, promote, edit mid-run (any agent role)
💾 Docs Live markdown memory
⚙️ Settings Provider, knobs, safety

Mid-run editing ✏️

While agents run you can drag cards, promote columns, edit CONTEXT/MEMORY, and add notes. The loop reloads board.json each wave. Chaos, but structured chaos.

Human-in-the-loop modals ✋

Studio blocks the relevant step with a modal (pipeline header shows Awaiting you):

Modal When Options Timeout default
Clarify vague PRD interview pick options 2m → recommended
Plan approve before execute approve / replan 2m → approve
Escalate task hits max review retries re-scope / retry / mark done / abort 30s → @escalate SLM decides
Continue QA/retries exhausted continue / stop / flag 2m → stop
Shell shell_permission=ask approve / deny 2m

Config (Settings → Planning / scope, or YAML):

escalate_ask: ask              # ask | auto | off
escalate_ask_timeout: 30s      # then @escalate (or escalate_timeout_agent) decides
escalate_timeout_agent: ""     # empty = @escalate → @reviewer → @coordinator
continue_ask: ask

TUI: /escalate re_scope|retry|mark_done|abort while the banner shows escalate pending. On timeout the dedicated @escalate arbitrator picks retry / re-scope / abort / mark_done.

API: GET /api/escalate/pending · POST /api/escalate/answer {"action":"retry"}.

Live events (SSE) 📡

Same stream as slmcode run -v:

Field Example
agent @worker
kind agent_start / coord / learn / output
task_id T1
scope focus files
output truncated specialist text 
curl -N http://127.0.0.1:7420/api/events

HTTP API 🔌

Method Path
GET /api/health
GET/PUT /api/config
GET/PUT /api/pipeline · POST /api/pipeline/reset
GET/PUT /api/docs, /api/docs/{name}
GET/POST/PATCH/DELETE board / tasks
GET/POST/PUT/DELETE /api/agents (custom + overrides)
GET /api/skills /api/models
POST /api/runs /api/runs/stop
GET /api/runs/latest /api/events
GET/POST /api/escalate/pending · /api/escalate/answer
GET/POST /api/continue/pending · /api/continue/answer
GET / SPA

→ Full pipeline schema: Pipeline

curl -s http://127.0.0.1:7420/api/health | jq .
curl -s -X POST http://127.0.0.1:7420/api/runs \
  -H 'Content-Type: application/json' \
  -d '{"query":"Add a doc comment to Hello()"}'

Pair with the TUI 🥊

# A
slmcode studio
# B
slmcode watch

→ 🖥️ TUI · 🧪 Recipes

☀️ Made with ♥ by UnicoLab