Cohere Reference Architecture
Overview
Define clear boundaries between clients, policy, retrieval, provider access, tools, evaluation, and operations so model behavior cannot bypass application controls.
Prerequisites
- The target repository, runtime, environment, and accountable owner
- An approved Cohere team and key for any live verification
- Current quality, security, privacy, capacity, and change-control requirements
Tool Discipline
Use Read, Glob, and Grep to inspect code, configuration, and evidence. Use WebFetch only for current Cohere primary documentation. Use Write or Edit only when the user requested implementation and the exact target files are known; never write credentials or customer content.
Current Contract
- Keep a typed Cohere provider adapter behind an application service boundary.
- Apply tenant authorization and metadata filters before retrieval, then preserve document IDs through Rerank and citations.
- Dispatch tools through a policy-enforcing registry with schemas, approvals, timeouts, and audit evidence.
- Resolve model, lifecycle, capacity, and platform IDs through environment configuration and current provider sources.
Authentication
Use an environment-specific key injected from an approved secret manager. Never print, persist, commit, or place CO_API_KEY in an example. Confirm access with the least costly bounded operation appropriate to the task, and treat key creation, rotation, revocation, role changes, and production-capacity requests as owner-approved actions.
Instructions
- Capture users, tenants, data classes, quality objectives, latency targets, side effects, and failure tolerance.
- Draw trust boundaries for clients, API, policy, retrieval, vector store, Cohere, tools, logs, and secret stores.
- Define request, result, citation, tool, error, and telemetry contracts at each boundary.
- Choose synchronous, streaming, queued, or degraded paths with bounded retries and cancellation.
- Attach offline evaluation, protected live probes, canarying, observability, budget, and incident controls.
- Review threats, data terms, capacity, model lifecycle, rollback, and named ownership before implementation.
Approval Boundaries
Do not expose or rotate keys, change Cohere Team roles, accept commercial terms, enable sensitive production data, increase spend or capacity, switch production models, send a support bundle, or execute model-proposed side effects without the accountable owner's approval. Keep diagnosis read-only unless implementation was requested.
Output
Return the resolved API and model contract, files or settings inspected, evidence collected, validation result, remaining risk, owner, and rollback or next action. Redact keys, authorization headers, prompts, retrieved documents, embeddings, customer identifiers, and unrestricted environment output.
Error Handling
| Condition | Response | |---|---| | Provider calls in UI | Move them behind the trusted server boundary. | | Tenant filter after retrieval | Move authorization before any candidate can cross the boundary. | | Tool reflection | Replace it with an explicit validated registry. | | Implicit fallback | Require configured, evaluated, and approved model routes. |
Examples
Use this compact handoff shape to keep the selected scope, validation evidence, and operational result reviewable.
Input:
system=enterprise-search; tenants=multi; features=rag,tools,streaming
Expected handoff:
boundaries=defined; contracts=versioned; controls=owned; rollback=designed