One platform.
From connected to running in production.
Sempleo is the context layer, the agent engine, and the workspace where agents run — one continuous system, not three integrations. Every piece is built to be auditable, reviewable, and yours.
Every capability,
on the same spine.
Five-layer context model
Company, team, client, project, and user — curated, versioned, and reviewable.
- Ownership per layer
- Freshness per field
- Full audit trail
OAuth connectors
Gmail, Calendar, Drive, Slack, HubSpot, Jira, and Notion today. Hybrid retrieval against your vector store.
- OAuth per workspace
- Per-tenant embeddings
- Reference, don’t copy
Install, don’t train
Roughly thirty agents across eleven categories, each carrying a Conversational, Hybrid, or Automation badge. Install per-team or per-workspace; fork any of them; publish your own as private catalogue entries.
- ~30 starter agents
- 11+ categories
- Install, fork, or publish
Workflows in plain English
The Workflow Builder is an installed agent. Describe the workflow in plain English; it parses intent, maps your installed agents’ input/output schemas, drafts the definition, and previews it for confirmation.
- Conversational authoring
- Multi-step branching
- The spec is the artefact
Context health dashboard
Per-field quality scored by Haiku, rolled up per layer and across the workspace. Where a field is thin, “Assist” proposes a fill — a human approves.
- Per-field Haiku scoring
- Workspace + layer rollups
- Per-agent readiness
Agent runtime
Every agent is a readable spec with declared context requirements, tools, and a rubric. The runtime resolves context, runs the agent, and writes a per-run trace.
- Versioned, reviewable
- Context-aware by design
- Rubric-gated outputs
A review queue, not a chat
Writes land as pending review — the agent does the drafting, your operator approves the send.
- Per-run rationale
- Cited evidence
- Approve, edit, or reject
Rubrics + audit log
Every output scored against a per-agent rubric. Every context change and every run logged; retention scales with plan.
- Per-agent rubrics
- Immutable audit log
- Scoped API keys
Per-run trace
See which layers attached, which sources were cited, how long each step took, and what it cost — every run.
- Layer attachment view
- Citation list
- Latency & cost breakdown
Open by protocol
Speaks the Model Context Protocol both ways. Sempleo is an MCP server for any client that reads it, and an MCP client for your own servers.
- MCP server · read + suggest
- Up to 3 external servers (Pro)
- Unlimited (Enterprise)
Bring your own model
Run on Sempleo’s managed default, or route through your own Anthropic or OpenAI account on Enterprise.
- Sempleo-managed default
- BYO Anthropic or OpenAI
- No training on your data
Four moves,
kickoff to production.
Connect the tools
you already use.
OAuth into Google, Atlassian, HubSpot, Slack, and Notion. Point at any MCP server you already run. Sempleo references your data with per-layer access control, not a shadow copy.
Reference, don’t copy. Sempleo reads from your source of truth. Revoke access and the context evaporates — no shadow copy.
Seed the
five layers.
Agents propose context entries from your existing artefacts. A human reviewer stamps them verified.
propose client.rebrand.tone # from 12 sent threads · confidence: high
propose team.vocab.sprint # from 6 standups · confidence: medium
propose user.signoff # from 40 sent · confidence: high
Install agents.
Don’t train them.
Marketplace of roughly thirty starter agents across eleven categories. Each carries a Conversational, Hybrid, or Automationtype badge. Install per-team or per-workspace, fork when a starter doesn’t fit, or author your own through the Workflow Builder — in plain English.
Email Drafter
CommsOutbound email in your voice, grounded in thread history and client tone.
SOW Drafter
SalesAssembles statements of work from the brief and your firm’s commercial template.
Account Researcher
ResearchPre-call brief from CRM, news, filings, and prior engagement notes.
Project Status
DeliveryWeekly status from Jira, commits, and Slack — written in your house format.
Every installed agent carries its context requirements explicitly. If a layer is missing, the agent surfaces the gap rather than silently guessing.
Trace every run.
Review what matters.
Every agent run shows which layers attached, which sources were cited, how long each step took, and what it cost. Flag, rerun, rollback — all native.
Nothing ships without a human unless you say it can.Low-stakes tasks auto-send when voice-match > 90%; sensitive tasks always queue for approval.
Sempleo speaks MCP
both ways.
Your team context, over the protocol every modern LLM speaks.
Point Claude Desktop, Cursor, Windsurf, or any other MCP-aware client at your Sempleo workspace. The client reads the same five-layer context your agents do — scoped to the user, authority-respecting, and fully audited.
- Read tools · context, knowledge, runs
- Write tools · land as pending review
- Per-user OAuth · nothing leaks across seats
Your own MCP servers, live at retrieval time.
Register the MCP servers you already run — internal warehouses, vertical SaaS, whatever your team has exposed — and Sempleo calls them live when an agent runs. Circuit breakers and rate limits are built in; the audit log captures every call.
- Up to 3 external servers on Professional
- Unlimited on Enterprise
- Per-tool allowlist · scoped credentials
Sempleo lives
where you work.
The workspace, full-fat.
Admin, review queue, context editor, per-run trace — the whole surface, in every modern browser.
A native app on macOS and Windows.
Built on Electron. Keeps a persistent review queue in your tray so agent drafts find you, instead of sitting in a tab you forgot to open.
Approvals in your pocket.
React Native app for iOS and Android. Review and approve agent runs from wherever you actually are when the queue fills up.
A platform,
not a point tool.
Day-1 connectors
for the systems your operators live in.
Most “AI platforms” are three products stitched together: a connector, a model, and a wrapper. I’m building Sempleo as one continuous system — context, agents, and workspace on the same spine — because the seams between those layers are where trust breaks. If you can’t read what the agent ran on, you can’t ship it into a regulated workflow.
