Expert Crews

Your best experts, available to every engineer. As agents they can ask from their own tools.

The DevOps team knows your infrastructure. The product team knows the requirements and how features are tested. Today everyone waits for them on Slack. With Expert Crews, each team builds a specialist Crew with its knowledge, skills and tools, and shares it. Any engineer can ask it from their own Claude Code, Codex, Cursor or ChatGPT.

  • 01
    Experts are the bottleneckThe same infrastructure, database and product questions reach the same few people every day.
  • 02
    Knowledge stays in heads and threadsRunbooks, requirements and test plans are scattered, so every answer starts from scratch.
  • 03
    Every engineer builds their own agentPeople paste context into personal chats and prompts, and no one shares or improves them.

Example goal

Expert answers without the wait

Primary metricQuestions answered by an Expert Crew
70%

answered without a human, target 60%. Illustrative numbers.

How it works

Built once by the experts. Used by everyone.

  1. Experts build the agent

    The DevOps team sets up a Crew agent with its runbooks, skills, memory and read access to the right systems. The product team does the same for requirements and testing.

  2. Offer clear functions

    Besides free-text ask, a Crew can offer typed functions such as diagnose_rds(instance) or release_checklist(feature), with inputs and results checked.

  3. Share it with the teams

    Give people access with scoped tokens. They see what the agent offers, never its private chats, data or settings.

  4. Ask from any tool

    Engineers call it from Claude Code, Codex, Cursor, ChatGPT or Claude through MCP, or with agentworks crews ask. Each person gets their own conversation with it.

What you get

An internal library of expert agents. Owned by the teams who know best.

Examples teams build first. Each is a Crew agent the owning team keeps improving, while everyone else just asks.

  • DevOps agent: deploys, pipelines, on-call runbooks
  • Database agent: RDS and Postgres health, slow queries, safe changes
  • Product agent: requirements, acceptance criteria, how a feature is tested
  • Security agent: policies, approved libraries, review checklists
  • Data agent: where a metric comes from and how to query it
  • Onboarding agent: how the codebase, services and tools fit together

Built for your security review

Runs in your cloud. Every action on the record.

  • Self-hosted

    Deployed in your cloud account or data center. Evidence stays in your environment.

  • Approvals

    Anything that changes production or reaches people waits for approval by default.

  • Audit trail

    Every run, tool call, decision and cost is recorded per workflow.

  • Your models

    Your enterprise Claude, ChatGPT or Gemini agreements, or private endpoints.

  • Scoped access

    Tools, folders and secrets are granted per workflow, inside an OS-enforced sandbox.

  • SSO and roles

    Sign-in through your identity provider, with roles and per-workflow access.

FAQ

Questions, answered.

How do engineers reach a shared agent?

Through MCP from any client that supports it, such as Claude Code, Codex, Cursor, ChatGPT or Claude, or with the agentworks CLI. They can ask in free text or call its typed functions.

Can people see the agent's private data?

No. Callers get the agent's answers and the files its owners choose to share. Its chats with owners, its database and its settings stay private.

Who controls access?

Admins issue scoped tokens per person and per agent. Revoking a token stops its calls immediately, and every call is recorded.

Does every caller share one conversation?

No. Each caller gets their own continuing conversation with the agent, and the owners' main chat stays separate.

Pick one expert to share first. See it answer in week one.

We build your first Expert Crew with the team that owns the knowledge, then connect it to your engineers' tools.