Account and workspace scope follows every customer action.
Human membership, service principals, tokens, runs, connections, and approvals are admitted against explicit account and workspace authority.
Governed AI security
Useful autonomy starts with explicit authority.
Colter places account and workspace scope, capability grants, tool policy, approvals, budgets, checkpoints, events, and attributable outcomes around durable AI work. Greater autonomy should make control more visible—not less.
Control around the run
Security decisions become part of the workflow design.
Every production proposal identifies the source systems, data boundaries, model routes, credentials, tool actions, approval points, writeback paths, and people accountable for the result. Provisioning follows that accepted boundary.
Colter is currently in private beta. Security posture, support expectations, retention, and production-readiness evidence are reviewed with each customer before activation; this page does not claim certifications Colter has not earned.
Current control model
Scope identity, context, action, and evidence together.
Agents receive bounded models, context, tools, and budgets.
Profiles and installations define what can run; grants classify read and mutation authority before a tool call reaches execution.
Consequential actions can stop for review.
Approvals, checkpoints, appended input, cancellation, and resumable state keep people in control without discarding completed work.
Events, usage, tool activity, and outcomes stay attributable.
Runs maintain durable identity and recorded history so operators can inspect what happened, recover safely, and explain the resulting state.
Before production
We document the operating boundary with you.
The accepted proposal and technical kickoff identify data handling, model providers, integration credentials, approval policy, usage limits, support ownership, evaluation evidence, and the exact path by which results return to the application.
See the review processSecurity contact
[email protected]
Send security questionnaires, architecture-review requests, privacy questions, or responsible disclosures directly to the Colter team.
Email the security contactReview before access
Bring security into the first workflow conversation.
We’ll make data, model, tool, approval, and writeback decisions explicit before provisioning.