Skip to content
SpaceKraftOS
DocsSign In

Agent operating system · v0.9.1

Agents thatdeliver results.

SpaceKraftOS runs your AI agents in one governed portal. You build them, approve what matters, and can prove every step. Use it as software, or let it deliver an agreed result for you.

Open the PortalRead the Docs

Runs Today

1,840

+12% on yesterday

Audit Chain

Verified

61 entries, no gaps

Waiting for You

High Risk

Incident Responder wants to page the on-call rota and post a status update.

ApproveReject
Example screens from the portal.
consoles in one portal
5consoles in one portal
roles, from guest to break-glass
25roles, from guest to break-glass
permissions, each reviewed in code
73permissions, each reviewed in code
audit chain for every action
1audit chain for every action

One portal,five consoles.

Every role sees the console it needs and nothing it shouldn't. Your roles decide which doors open, and the server checks again on every action.

  • Console 1

    Platform Admin

    Global cluster telemetry, LLM provider health, edge routing, and the emergency kill-switch.

    For platform operators and sres

  • Console 2

    Tenant & Org Admin

    SSO, SCIM provisioning, RBAC role mappings, VPC scope, MCP enterprise tools, budgets and spend.

    For organisation admins

  • Console 3

    Builder & Operator

    Natural-language agent compiler, graph blueprint editor, agent fleet, and the HITL approval gateway.

    For agent builders and operators

  • Console 4

    Governance & Audit

    Immutable Merkle audit stream, policy enforcement engine, and compliance posture (SOC2 / HIPAA / EU AI Act).

    For auditors and compliance

  • Console 5

    End-User Surface

    Verifiable assistant with step-by-step reasoning traces, DID mandates, and DLP redaction.

    For everyone who works with agents

  • One sign-in, every console your roles allow.

    No hidden admin: even the break-glass role is audited.

    Open the Portal

From idea to proof,in four steps.

  1. Build

    Describe the job. The compiler turns it into an agent graph you can read, edit and version.

  2. Govern

    Policies and human approvals live in the graph, so risky steps wait for the right person.

  3. Run

    Agents use your tools through MCP, remember with tiered memory, and stay inside their budget.

  4. Prove

    Every action lands on a hash-chained audit stream that anyone can verify.

Run it yourself,or get the result.

Two ways to work with SpaceKraftOS, on the same platform, with the same approvals and the same audit trail.

SaaS · Software as a Service

Your team builds and runs the agents.

  • Your own organisation, with single sign-on and role mappings
  • The builder, the approval queue and the agent fleet
  • A visual editor for your real agents: save, publish, run live and replay, or simulate on sample data
  • A live map of your agents at work, schedules, and catalog agents that update when we improve them
  • Your own Python or Node.js functions as agent tools, and sandboxed JavaScript code steps
  • People and roles managed in the portal: invite, change roles, suspend, every change audited
  • Budgets and spend per organisation
  • Your tools connected: GitHub, Slack, Jira or any MCP server
  • Alerts in Slack, Teams or your own webhook, and API keys for your own systems
  • A website scanner for SEO, AEO and GEO, with fixes to paste, and InfiScan to run it from a Windows PC
  • HireWise: post a job, screen CVs and interview by text, voice or video with AI, with quoted evidence and abuse checks; a person makes every decision
  • DineWise: take restaurant bookings by web form, chat, voice, phone and message with one AI agent, in the languages each location serves

Best when you have a team that wants to build.

Open the Portal

RaaS · Results as a Service

You agree the outcome. Agents deliver it.

  • An agreed result with clear acceptance criteria
  • You approve the steps that matter, nothing more
  • Every result arrives with its full audit trail
  • Built and run on the same governed platform

Best when you want the result, not another tool.

How RaaS Works

Governed by design,not by promise.

The controls sit below the application, where a bug or a forgotten check cannot switch them off.

  • Tenant isolation in the database

    Row-level security scopes every query to one organisation. A query that forgets its filter returns nothing.

  • Roles you can read

    25 roles from one reviewed lattice. Platform roles cannot read customer data without a grant, apart from two named cross-organisation roles.

  • Human approvals

    High-risk steps pause until someone with the right role approves, and the decision is recorded.

  • Emergency kill switch

    Halts every agent in every organisation at once. Reversible, restricted and audited.

  • Verifiable audit chain

    Append-only in the database and hash-chained, so a changed or missing entry shows.

  • Compliance mapping

    Controls mapped to SOC 2, HIPAA and the EU AI Act, in one posture view. A map, not a certificate.

See itrunning.

Sign in to the demo portal, or start with the docs: the architecture, the API and the release notes.

Open the PortalRead the Docs