The future of engineering intelligence shouldn't come at the expense of engineer privacy
A new era of visibility calls for a new standard of trust. Here’s how we’re building both into Span.
Privacy-first by design
Agent traces represent an extraordinary opportunity. For the first time, organizations can understand not just what gets built, but how it gets built. What's working, what's not, where agents struggle, and where teams can improve.
But traces can also contain some of the most sensitive information inside an organization: proprietary code, business plans, personal conversations, and the messy, exploratory process of solving hard problems.
We believe organizations should be able to learn from this data without compromising the trust of the people creating it.
That's why we've made privacy a foundational part of how Span works.
Control what's collected
Not everything belongs in a trace. Configure collection policies, including specific repositories and file paths, to limit what enters the system in the first place.
Protect data at the source
Span applies redaction locally, on the developer's device, before events are transmitted. Secrets, credentials, and personal information are scrubbed before they ever reach the network.
Redact again in the cloud
Before anything is stored, a second independent pass uses named entity recognition and zero-retention ML to maximize detection rates and prevent inadvertent secret or PII capture.
Look beyond keywords
A job search, performance discussion, or confidential planning session may contain no obvious secret. LLM-based topic screening detects sensitive traces and quarantines them for review before inclusion in analysis.
Keep individual traces private
By default, an engineer's prompts and full session contents are visible only to that engineer. Aggregated reporting lets teams improve without defaulting to individual surveillance.
Make policies configurable
Organizations have different expectations and requirements. Span supports privacy modes, granular roles, and permissions mapped to people, teams, and organizational structures.
Your trace belongs to you
Trace-level insight and organization-level learning don't have to mean unrestricted access to everyone's work.
Private by default
By default, raw prompts and trace conversations are restricted to their author. Separate permissions govern metadata and insights. Organizations can choose alternate policies where appropriate.
Granular access
Role-based permissions can scope trace metadata, reporting, and management to appropriate people and teams.
Audited controls
Audit logs, identity management, encryption, and data retention policies support enterprise governance.
Flexible deployment options
Customer-controlled cloud deployment choices for organizations that need them.
