Early access · Speakeasy Assistants

Assistants that do the work, enterprise ready

Run on the Speakeasy AI control plane. Every assistant connects through the MCPs and Skills your org already uses, with identity, guardrails, and audit built in.

Speakeasy Logo
Speakeasy AssistantLive
Speakeasy Logo

What should we get done?

Connect to your tools. Describe a workflow. Watch the assistant execute it across systems, with full audit and control.

Tools, identity, and audit logs handled by Speakeasy.

↑ Try it: type a workflow or click a starter

Powered by the AI control plane

Not another chatbot. The work, run on the plane you already trust.

Speakeasy Assistants run on the same AI control plane that connects, controls, secures, and observes every other AI client in your org. Every workflow inherits the MCPs, Skills, identity, and audit trail you already trust.

Read the reference architecture
  • Connect
    MCPs, Skills, SaaS, internal APIs
  • Control
    IdP-bound identity, RBAC, OAuth 2.1
  • Secure
    PII redaction, prompt injection, shadow MCPs
  • Observe
    Audit trail, token use, adoption

  • Connect: MCPs and Skills, on tap

    Assistants reach SaaS apps, internal APIs, and databases through the same MCP servers and Skills that power the rest of your AI stack. Build once, share across every team and every assistant.

  • Control: identity from your IdP

    RBAC and OAuth 2.1 inherit from Okta, Entra, Auth0, or whichever IdP you already run. Scope assistants by team, role, or tool. No API keys pasted into config files.

  • Secure: guardrails on every call

    Actively block PII and data exfiltration. Passively detect prompt injection and shadow MCPs. Every prompt, response, and tool call flows through one inspectable plane.

  • Observe: full audit and adoption analytics

    Every tool call is logged with a full audit trail. Token use, adoption, and tool usage broken down by team, client, and assistant. Data you can hand to security and the board.

  • Multi-step workflows, not single-shot prompts

    Assistants reason across tools, pulling data from one system, transforming it, and acting in another. The control plane handles orchestration, retries, and context.

  • Built for the whole org

    Share assistants with a team, scope them to a workflow, or roll them out to the whole company. SSO and SCIM included. Same plane every other AI client already runs on.

How it works

Four steps from idea to a shipped workflow.

Each step maps to a control plane primitive your security team already signed off on.

  1. 01
    Connect

    Wire up MCPs and Skills

    Browse the catalog, point at an OpenAPI spec, or attach an existing Skill. Every tool becomes a managed MCP server the whole org can share.

  2. 02
    Build

    Define the assistant

    Pick a goal, scope the toolset, set guardrails. The control plane handles model orchestration, retries, and context.

  3. 03
    Control + Secure

    Roll out safely

    RBAC and OAuth 2.1 inherit from your IdP. PII is redacted in flight, prompt injection and shadow MCPs are detected at the plane.

  4. 04
    Observe

    Watch and measure

    Inspect every tool call, prompt, and response in real time. Full audit trail plus adoption and token metrics by team, client, and assistant.

Use cases

One assistant. Every team.

Engineering

Ship faster, on-call less

Triage incidents, draft RFCs, and move tickets without leaving the assistant. Connect Datadog, GitHub, Linear, and PagerDuty in minutes.

Revenue

Close the loop on every account

Pull pipeline data, draft outreach, and brief reps before every call. Salesforce, HubSpot, Gong, and Slack, all in one workspace.

Operations

Automate the boring middle

Reconcile, report, route. Assistants chain reads and writes across your finance, HR, and ops stack with full audit logs.

Start with Assistants

Give your team an assistant that actually does the work

Try AssistantsBook a demo