Use case
A Read-Only Trust Layer for Agent Writes
Agents that write to external systems need a trust layer they can query without risk.
If the trust layer itself writes, it becomes another source of data quality problems.
Anchord evaluates proposed writes without side effects — your agent gets
allowed or
blocked, and Anchord
never touches the external system.
Why read-only matters
A trust layer that writes to external systems creates a new vector for data quality problems. If the identity resolution service creates records in Salesforce, it can create duplicates. If it updates records in HubSpot, it can overwrite good data. The tool meant to prevent bad writes becomes capable of bad writes itself.
Anchord avoids this entirely. It ingests data (reads from external systems via OAuth or API), resolves identity, and evaluates proposed writes. It never creates, updates, or deletes records in any external system. The separation is architectural, not just a configuration option.
Guard semantics
The guard endpoint evaluates a proposed write and returns a decision. It does not execute the write.
allowed: true The proposed write is safe to proceed. The entity exists, links are consistent, and no blocking conditions were found. Your agent executes the write to the external system.
allowed: false + blocks array
The proposed write should not proceed. The blocks
array contains specific codes explaining why — missing entity link, conflict with another
record, or other blocking conditions. Your agent reads the codes and decides what to do.
Trust without side effects
Every guard call produces an audit_id.
Every resolve call produces an attempt_id.
All decisions are logged and reviewable in the Anchord console.
The audit trail answers questions after the fact: what did the agent try to write? Did Anchord allow it? Why was it blocked? Which entity was resolved? This is the transparency that production agent workflows require — without the risk of the trust layer itself causing damage.
Your agent calls Anchord, gets a decision, and acts on it. Anchord's job is done after returning the response. It has no access to write to your CRM, your billing system, or any other external system.
Evaluate writes without risk
A trust layer that never writes. Query it freely — the answer is the only output.