Durable agents.
Fully managed.
The control plane your product needs before it can ship agents — identity, memory, sessions, budgets, and governance. One API. No side projects.
Hosted in Skopik Cloud. Reaching any hardware you own.
// Create a durable agent
const agent = await client.agents.create({
name: 'Atlas',
instructions: 'Track pricing. Cite sources.',
dailyBudgetUsd: 5,
})
const session = await client.sessions.open({
agent: agent.agentId,
prompt: 'Compare our top competitors.',
})
// Reconnect without losing progress
const events = client.sessions.stream(session.sessionId, {
after: cursor,
})
for await (const event of events) {
render(event.envelope)
}Persist first, then publish. Reconnect from any cursor.
The platform primitives
8 nouns, no more. Everything else is a session field, a tool, or metadata.
Agent
A durable identity. Name, brain, model, budget, history — created once, never rebuilt from a prompt.
Session
That agent doing work. Prompt, steer, stream, replay — one record that survives deploys and idle weeks.
Remote
Hardware a session can reach. Work folder + tools on your laptop or server — loop stays in the cloud.
Every agent product
rebuilds the same
infrastructure
Your agent is a process in a container. The container lives until the next deploy — and everything the agent was doing dies with it.
So you bolt on Redis, a queue, a cron sweeper. Now you're maintaining a distributed-systems side project instead of your product.
That whole layer, built once, operated for you.
- Session store that survives deploysDurable transcript, files, budget, lineage — one record from first prompt to final result.
- Resumable streaming that actually worksPersist before publish. Reconnect from cursor, replay, rejoin live.
- Budgets that are hard limitsWarnings, then new work stops. Every cent attributed to org → user → agent → session.
- Governance from day oneScoped keys, approval gates, append-only audit.
Compute moves. Agent identity does not.
Day-to-day
What this looks like in practice
Seven things that stop being problems the moment you put them behind a control plane.
Deploys don’t interrupt work
Agent loop isn’t in your process. Ship whenever you want — sessions keep transcript, files, and budget.
Disconnect without losing anything
Every chunk persisted before it’s published. Reconnect from cursor and replay what you missed.
Steer a session while it runs
New instructions land in a durable inbox and merge at the next safe boundary. No race conditions.
Edit behavior in plain files
Brain is one live file tree. Add a rule, adjust tone — no deploy, next session picks it up.
Budgets are hard limits
dailyBudgetUsd: 5 means $5. Warnings first, then new work stops at the ceiling.
Every channel reaches one session
Slack, email, schedules, and API calls open or rejoin the same session — one record of work.
Work runs on your hardware when needed
Bind a Remote and work/ lives on your machine. Reasoning and credentials never leave the cloud.
Also included
Templates and Skills for reusable agents, hybrid search over brain and memory, child sessions that delegate real work, and durable callbacks that park long waits.
The stack you delete
Your roadmap, minus the platform work
This is what you don't build, don't operate, and don't get paged for.
- Session store that survives deploys and restarts
- Resumable SSE streaming with cursors and replay
- Mid-run steering without race conditions
- Agent memory + hybrid retrieval pipeline
- Per-agent budget metering with hard stops
- Approval gates in front of dangerous tool calls
- Encrypted credential store for integrations
- Append-only audit log with attribution
Meet it where you work
Typed clients for agents, sessions, files, streams — plus a useChat-compatible transport.
npm i @skopiklabs/sdkOpen, steer, stream, approve, and cancel sessions from the terminal. Also runs a headless Remote.
npm i -g @skopiklabs/cliServe the whole platform over MCP — Claude, Cursor, or anything that speaks the protocol.
Streamable HTTP · stdioOne REST + SSE API in front of all of it.
SDK, CLI, and MCP all speak the same durable session protocol. No separate stores to reconcile.
Governed by default
Control that stays
with every session
Budgets, approvals, and audit aren't bolted on. They travel with the session from first prompt to final result.
Budgets with hard ceilings
Spend attributed by org, user, agent, and session. New work stops at the cap — in-flight finishes clean.
Approvals before consequences
Sensitive tool calls wait, durably, for a human yes. No timeouts, no lost state.
Scoped keys & postures
Read, write, and admin keys. Per-connector write posture, read-only Remotes.
Audit that can’t be edited
Every state change appends who did what, to what, in which session. Append-only.
Running agents somewhere regulated? Let's talk Enterprise.
Contact salesPricing
Start free. Grow without replatforming.
Free to start, Pro when early access opens, Enterprise shaped with you. Full limits on the pricing page.
Free to start
Skopik Free
1 seat · 3 agents · 3 connectors
Explore the control plane. No card required.
Join the waitlist- 50 sessions / mo
- 1 GB storage
- 100k in + 50k out tokens
For production teams
Skopik Pro
Target early-access pricing
Higher limits for production workflows.
Join the waitlist- 25 agents & 25 connectors
- 1,000 sessions / mo · 10 GB
- 5M in + 2M out + $20 calls
Built around your workload
Enterprise
Organization-specific terms
Annual commitments, custom allowances, and billing policy.
Contact us- Custom agents, seats, storage
- Dedicated BillingPlan
- Configured with Skopik staff
Pro pricing and allowances are indicative during early access. See full comparison →
Give your agents
somewhere to live
Create one agent, hand it real work, and close the laptop on it.