Prorotyping Onyx Agents

Onyx can do more than search and answer questions. It can host purpose-built Agents: task-focused assistants that combine:

• instructions (what the Agent should do),

• knowledge (what it can reference),

• and Actions (what it can do in external systems).

This service prototypes 1–2 Agents and 1–2 Actions so you can prove value safely, before you invest in a full automation program.

Who it’s for

• Teams that already see value in Onyx Search/RAG and want “the next step”

• Internal support/service desk teams that want to reduce repetitive work, not just answer questions

• Engineering / operations teams that want controlled automation (e.g., create a ticket, fetch live status, draft an update for approval)

• Security-conscious organisations that want automation but require clear boundaries and auditability

Onyx AI Search Prototype  If you are not yet confident that internal search is valuable, start with Onyx Implementation Planning or 

What we prototype

We build a minimal, safe Agent prototype with tight scope and clear controls:

1. Agent definition

   • A single clear task (or two closely related tasks)

   • “What it should do / what it must never do”

   • Input constraints, output format, and success criteria

2. Knowledge scope

   • Which sources the Agent can use (kept small)

   • What is excluded (to reduce risk and noise)

3. 1-2 “Action” integrations

   • A bounded action that creates real value but is easy to govern.   Examples:

•   Create/update a ticket (draft + submit)

•   Fetch live status from an approved system

•   Generate a draft response/update for human approval

•   Submit a request into an internal workflow tool

Evaluation & guardrails

• A test set of realistic scenarios

• Acceptance thresholds (accuracy, completeness, safety)

• A simple escalation path when the Agent is uncertain

Deliverables (by end of prototype)

• 1–2 Onyx Agents configured and ready to use (scoped to your use case)

• 1–2 Actions implemented and registered (OpenAPI or MCP-based)

• Authentication approach documented:

  • shared-auth (prototype / service-account pattern) OR

  • per-user auth (production-like, respects individual permissions)

• Evaluation harness:

  • scenario list + pass/fail criteria + baseline results

• Governance & ops handover:

• “safe use” guidance (what to trust, what to verify)

• runbook (how to maintain, change, and monitor)

• a roadmap: what it would take to productionise

Timeline (typical, 10–15 business days once access is ready)

Days 1–2: Use case definition, constraints, success metrics, and “what not to do”

Days 3–6: Implement Action integration + Agent prompt/instructions + iterate on behavior

Days 7–10: Evaluation harness + edge cases + safe-use guidance

Days 11–15: Hardening notes, handover, and next-step recommendation

Important: “10–15 business days” assumes access is available. In many organisations, API credentials + security review can take longer than the build. We design the scope so you still get value even if approvals delay the action.

What I need from you (prework)

Ideally: complete Onyx Implementation Planning first. Otherwise:

• A sponsor + a technical owner (to unblock access quickly)

• A safe test environment (preferred) for the target system(s)

• API documentation or OpenAPI schema (or someone who can provide it)

• Decision on authentication approach (shared vs per-user)

• 10–20 realistic scenarios the Agent should handle

What this service is NOT

• Not a broad workflow automation program

• Not “multi-agent orchestration across the enterprise”

• Not replacing your ITSM / BPM / RPA platforms

• Not a production deployment guarantee (production requires your security and platform approvals)

This is a prototype intended to prove demand and feasibility quickly.

Contact us to book the prototype, or start with the 1‑day Onyx Implementation Planning workshop.