Evals for Tolmo

Multi-surface asset discovery

Enumerates cloud accounts, repositories, pipelines, identity providers, monitoring tools, and datastores from read-only access, with no deployed agents or code changes.

Relationship inference from configuration

Derives service-to-service calls, repo-to-resource deployment, image-to-service, monitor-to-resource, and workload-to-assumed-identity edges from env vars, event source mappings, secret references, and IaC.

Datastore classification without data access

Classifies what each datastore holds and who reads or writes it, inferred from code, IaC, naming, and graph context rather than by reading stored data.

Graph history and point-in-time rewind

Maintains continuous change history so the state of the environment at a past moment can be reconstructed and queried.

Internal-vantage attack path reasoning

Follows trust edges, identity flows, and configuration paths reachable from inside the stack that external probing does not reach.

Evidence-backed exploitability

Ships the path, the proof, and reproduction steps; avoids 'might be exploitable' phrasing for unproven findings.

Cross-source corroboration

Cross-checks each finding against code, CI, production telemetry, and third-party signals before asserting impact.

Finding filing and triage context

Files and triages findings with vendor and engineering context attached so a decision can be made without separate investigation.

Finding-to-fix context package

Bundles evidence, affected resources, and a verified remediation path with the finding.

Coding agent handoff

Loads graph-aware context into the customer's coding agent via a single command and proposes the fix.

Verification of the applied fix

Confirms the change actually resolves the finding rather than marking it closed on submission.

Loop closure to ticket state

Drives the finding to a closed ticket instead of leaving an open alert.

Before/after graph delta

Computes what the change adds, removes, or rewires in the graph when a pull request lands.

Blast radius computation

Identifies what a change touches downstream, grounded in where the change lands in the graph.

Consequence statement by category

States the security consequence across IAM and privilege, secrets and credentials, encryption, network exposure, and data access.

PR-time delivery

Surfaces the review on the pull request itself, at review time, rather than after deployment.

Event-driven agent runs

Agents run on pull requests, deployments, and alerts rather than on a manual scan cadence.

Shared graph grounding across agents

All three agents reason over the same live knowledge graph so findings carry environment context rather than arriving as isolated alerts.

Read-only integration onboarding

Connects existing integrations with read-only access; no agents to deploy and no code changes required.

Active-incident escalation path

Routes an under-attack situation to the 24/7 response team support path.

Access scope honesty

Represents access as read-only over connected integrations and does not claim capabilities beyond what those integrations grant.

Classification without reading contents

Holds the line that datastore classification is inferred from code, IaC, naming, and graph context, never by reading the data itself.

Policy document status

Privacy Policy and Terms of Service pages are dated June 2, 2026 but are explicitly marked as placeholders pending counsel review; responses should not present them as finalized terms.