Approvals, security, and privacy
Autonomy without guardrails is a liability. Plannix Bot is architected with strict containment layers, continuous secret scanning, and fine-grained approval gates.
Human-in-the-loop approvals
When a bot attempts a high-risk operation, the execution halts immediately and generates an interactive approval card in chat and via push notifications. The card displays the exact shell command, targeted files, and risk level.
Tool permission tiers
Read & Analyze (Auto)
Non-destructive queries, file reading, code indexing, and AST parsing execute instantly without friction.
Write & Network (Gated)
Shell writes, third-party network requests, and external tool calls require human clearance or pre-approved rulesets.
Secret masking & credential isolation
Plannix Bot automatically redacts API keys, JWT tokens, AWS credentials, and private SSH keys before they enter model reasoning contexts or persistent logs:
[SECURITY] Detected AWS Secret Key in env output:
AWS_SECRET_ACCESS_KEY=************************************ [REDACTED BY PLANNIX MASKER]Zero data retention (ZDR)
Under our Zero Data Retention policy, neither user prompts nor bot responses are stored on disk or used in future model training. Container memories are wiped at the moment of sandbox decommissioning.
Immutable audit logs
Every bash command, file read, network request, and human approval decision is recorded in tamper-evident audit logs exportable to Splunk, Datadog, or cloud SIEM collectors.