Evaluate an AI data connector across source coverage, incremental sync, parsing, stable identity, permission propagation, tenant isolation, retrieval quality, provenance, agent contracts, observability, security and export. Require test evidence for your own data and failure modes—not only a connector logo.
Ingestion and freshness
- Supported objects, types and limits are documented.
- Initial and incremental sync are defined.
- Deletes, renames and permission changes propagate.
- Retries and backfills are observable.
- Stable IDs survive updates.
- Parsing failures are recoverable.
Permissions and security
- Tenant isolation exists at storage and retrieval.
- Source ACLs are enforced at query time.
- Credentials are encrypted and minimally scoped.
- Revocation latency is tested.
- Content is treated as prompt-injection risk.
- Audit events identify actor, source and outcome.
Read why metadata filtering matters before accepting vague “secure search” claims.
Retrieval and evidence
- Exact identifiers and paraphrases are evaluated.
- Filters cover workspace, source, type, time and access.
- Results include stable source identity.
- Snippets and locations support verification.
- Empty and conflicting evidence are honest.
- A repeatable evaluation suite exists.
Agent and operational fit
- REST or MCP matches the client strategy.
- Tool schemas, pagination and errors are model-usable.
- Usage, latency, errors and freshness are monitored.
- Breaking changes are versioned.
- Data and metadata can be exported.
- Deletion, retention and incidents have owners.
Score each item proven, partial, absent or not applicable and attach evidence.
Built for the next step
Evaluate SPYN3 against its real surface
Developer guides document current endpoints, permissions and source-aware response fields.
Open developer documentationSPYN3 is AI-ready data infrastructure for technical solopreneurs and AI builders. It connects business data into a workspace-isolated knowledge layer and exposes it through a unified REST API and hosted MCP server, with permission-scoped access and source-aware returns. SPYN3 is currently in private beta; current capabilities and product direction are documented on the product facts page.
Sources
- Microsoft Learn: document-level access control — permission metadata and query-time enforcement. Verified 13 Sep 2026
- Anthropic: prompt injection defenses — indirect prompt-injection risk and layered defenses. Verified 13 Sep 2026
- Anthropic: writing effective tools for agents — namespacing, response design and tool evaluations. Verified 13 Sep 2026