The agentic studio that runs our own products: parallel Claude Code sessions that start
already knowing your project — memory is injected at spawn, automatically.
Kanban → pre-seeded sessions, per-session costs, nothing irreversible without your click.
The helm, not the autopilot.
Open source Apache-2.0One-command installMemory 100% local (built-in RAG)Managed cloud 🇨🇭 from CHF 129/mo
Agent wrappers all look alike. SOKKAN does the one thing that compounds over time:
your project memory is automatically injected into every session at spawn —
starting a task is the “check your memory” ritual.
◈Auto-injected memory
Semantic recall (local RAG, ONNX embeddings — no external service) over the notes your sessions write. Every new session starts knowing what the previous ones learned.
❯The helm, not the autopilot
Multi-session chat built on the Claude Agent SDK: permission requests and questions become buttons. Nothing irreversible without your click. Nothing reaches Done without you.
▦Kanban → sessions
A card describes the task; ▶ spawn opens a pre-contexted session that proposes a plan and waits for your go. Per-session costs, an audit journal for every action.
One real take: kanban card → ▶ spawn → the session searches the project memory first,
recalls the port and the team convention — facts that only exist in the notes — grounds itself in the
code, then proposes a plan and waits for your go.
The Board: priorities, due dates, checklists — and on every card, ▶ spawn opens a session
pre-seeded with the task description. Every move lands in the audit journal.
From idea to production — we don’t drop you at deploy.
Agent editors stop at git push. SOKKAN operates production with you —
event-driven ops, human-gated: a production event doesn’t trigger a script, it spawns an agent
that has your project memory, and the go-ahead stays human — Telegram pings you wherever you
are. The loop closes — idea → prod → observation → idea.
Agentic observability
Prometheus + Grafana + Loki in your network. Your sessions read metrics and logs, and compose a dashboard on demand: « watch my API p95 and 5xx ».
Alert → diagnosis session
A production alert becomes an incident and starts a session pre-seeded with the metric, logs and memory, proposing a fix — and waiting for your go-ahead, with a Telegram ping so you can come give it. An on-call agent, supervised.
Secrets vault
Encrypted, injected into sessions as environment variables. The agent uses them to deploy without the value ever showing in the UI or going to the model.
HITL push
You launch and step away: when a session waits on your approval, it pings you (Telegram/webhook) with a link to come click.
Replayable runbooks
An ops procedure = a memory note. « Run » opens a guided session, step by step, human approval on anything irreversible.
Deploy & rollback
Ship an image to a fleet worker, roll back to the previous version in one click — versioned, supervised deployment, applied to your apps.
Nina, your DevOps aboard.
Operating a platform shouldn’t require a full-time ops person. Nina is SOKKAN’s
embedded DevOps assistant: she knows the product inside out — sessions, memory, fleet,
runbooks — and answers in the cockpit, right next to your work.
Typical conversations, the way they happen in the cockpit:
3:12 am — an alert becomes a diagnosis
NinaDevOps aboard · cockpit
The API has been throwing 500s for ten minutes. What do I look at first?
The api-5xx alert already opened a diagnosis session, pre-loaded with your project memory: it points at the payments worker OOMing since the 10:14 pm deploy and proposes a rollback. Nothing ships without your go.
⚑ incident #12 → session diag-api-5xx · waiting for approval
Memory that keeps you honest
NinaDevOps aboard · cockpit
My session wants to migrate to Postgres… didn’t we already settle this?
We did — the decision-sqlite-storage note, back in May: SQLite on purpose, single-writer service. Mark it priority: high and every new session will see it before proposing anything.
★ note pinned — injected at every spawn
The bill that stings
NinaDevOps aboard · cockpit
My AI credits melted this week. What happened?
Costs tab: on Tuesday the scraping-refactor session burned 4.2M tokens in a retry loop. Move that task to a small model and replay the rate-limit runbook — I’ll show you which one.
📉 −78% consumption the next day
The guardrail
NinaDevOps aboard · cockpit
Just give me the prod database URI, it’ll be faster.
No — I never read or hand out a secret. It lives in your vault, injected as an env var into the session: it can use it without ever displaying it. Not to you, not to me.
🔒 secrets: never shown, never logged
And Nina herself is under watch
Every request, every token of Nina is continuously measured at NINABOT — and whenever she touches your context, her inference is served from Switzerland (Infomaniak), mandatorily.
keep scrolling ↓
◈Aboard every SOKKAN Cloud instance — the 🧭 button in the cockpit. Included, zero setup, from your first login.
◈Swiss sovereignty — mandatory: whenever Nina has access to your context (your instance’s cockpit), her inference runs on Infomaniak’s AI API, hosted in Switzerland. Non-negotiable — and the content of your conversations is never retained nor used for training.
◈Self-hosting? She ships in the repo — enable SOKKAN_FEATURE_ASSISTANT with the model of your choice.
Questions about SOKKAN? Nina also answers right here — bottom right of this page. ↘
On your machine, in one command.
Docker + your Anthropic API key (or your Claude Pro/Max subscription). Everything stays on your machine.
$curl -fsSL https://sokkan.ch/install.sh | sh⧉
Requirements: Linux (or macOS + Docker Desktop), ~4 GB RAM, an Anthropic API key or a Claude Pro/Max subscription.
No Docker? The script offers to install it (official get.docker.com script).
Sources: tarball · script — read it before piping to sh, obviously.
Or operated for you, in Switzerland.
The same SOKKAN, minus the ops: a dedicated VM in the Geneva zone, your own private network,
your URL at you.sokkan.ch. Bring your Anthropic key (BYOK) or take managed inference:
Qwen3 Coder models served from Frankfurt (EU), prepaid credits
(from CHF 25, +25% credits from CHF 500), metered per token. And the fleet grows from the cockpit: worker instances and managed PostgreSQL
inside your network, two clicks away. What you deploy there ships in one click too:
an HTTPS subdomain on sokkan.ch, or your own domain (one CNAME, automatic TLS).
Free 14-day trial on every plan — card required, first charge on day 14,
cancel in one click before (trial seats are limited).
Grow from the cockpit: worker instances from 59 CHF/mo · managed PostgreSQL from 99 CHF/mo (daily backups),
all inside your private network. Web exposure included: HTTPS subdomains and your own domains, automatic certificates, at no charge.
Prorated billing, provisioning after payment, cancel monthly.
Hosted at Exoscale (Geneva), operated by NINABOT Sàrl — your data stays in Switzerland.
Not sure about sizing? Write to [email protected] — you'll get the founder, not a bot.
My fleet, in the cloud cockpit: your workers and databases, live status,
and the catalog to order more — the cockpit holds no cloud credentials; everything goes
through the NINABOT control plane, payment first.
SOKKAN is open source (Apache-2.0) — born as the internal cockpit running
ninjob.ch and the NINABOT products, extracted because
memory is the part that compounds. Self-hosted and BYOK: free, no limits.
Community repo: github.com/ninabot-ch/sokkan — the code itself also downloads right here.
The managed cloud pays for the operation, not for un-crippled features: same product, same repo.