Security
Runs on your hardware. Answers to your rules.
Reqursor Development is installed on your own server or Raspberry Pi. Here is exactly what it sends where, what it can touch, and what it never will.
Data flow
What leaves your infrastructure
Your code and tickets leave the machine for exactly three kinds of destination, all of them chosen and configured by you.
Your server or Raspberry Pi
Reqursor Development: the dashboard, the worker and the agent sandbox, plus your tickets, run history, audit log and encrypted secrets.
1
The model provider you choose
Prompts and working context, including the code the agent reads, go to Anthropic, or to Amazon Bedrock, Google Vertex AI or Microsoft Foundry if you use them.
2
Your git remote
Commits and pull requests go to the repository you connected, on the branches you allow.
3
The integrations you connect
Your tracker, email, Slack or webhooks receive the updates you switch on. Nothing more.
Reqursor: license check only
Once a day, eight fields of metadata. No code, tickets or names.
License check
Eight fields a day. Nothing else.
Once a day, your install checks its license with Reqursor. The request contains exactly these fields:
- No code or diffs
- No ticket text
- No repository names
- No user names or email addresses
If the license server can’t be reached, work carries on and a banner appears. A short outage never stops your team.
// POST, once a day{ "instance_id": "inst_7f3c…", "license_id": "lic_2a91…", "version": "1.0.0", "os": "linux", "arch": "arm64", "active_users": 4, "projects": 2, "runs_24h": 17}Agent isolation
The agent only sees the code it works on
Ticket text is untrusted input, so the agent runs with the least it needs.
- Runs in a sandbox that sees only the project’s working copy
- Never sees your data directory, your license or your real API keys
- Model calls go through a local gateway that adds the key on the agent’s behalf
- Pushes, pull requests and tracker updates are made by Reqursor Development, not by the agent
- Ticket text is treated as data, never as instructions, and scanned for prompt-injection patterns
Your server
Reqursor Development
Agent sandbox
Guardrails
Rules enforced in code, not in prompts
A model can be talked out of an instruction. It can’t be talked out of code that refuses the command.
No force-push
Force-pushes are refused outright.
No history rewrites
No rebase, amend or hard reset on shared history.
No skipped hooks
Commits that try to skip your git hooks are refused, so your hooks always run.
Protected branches
The agent never pushes to the branches you protect.
Protected paths
Changes to CI workflows, environment files or secrets folders refuse the whole commit.
Approval gates
Large or sensitive diffs wait for a person to approve them.
Secrets and access
Locked down by default
Secrets
- Encrypted at rest with AES-GCM
- Write-only through the API, shown only as ••••last4
- Redacted from logs, live events, the audit log and diagnostics
- Never sent to Reqursor
Access
- Sign-in required for every page and action
- Roles: Owner, Admin, Developer and Viewer
- CSRF protection and scoped API tokens
- Metrics endpoint protected by a token
Audit trail
Every action on the record
An append-only audit log records every run, decision and settings change. Export any run as a post-mortem in Markdown or JSON for your own records.
# audit log · platform-api02:14:07 run.started RD-142 autopilot02:31:52 check.passed RD-142 tests · 212 passed02:39:40 verify.agreed RD-142 4/4 criteria02:40:10 ship.approved RD-142 evidence complete02:40:11 commit.created RD-142 9f3c2e103:02:45 approval.held RD-141 412 lines changed08:05:33 decision.approve RD-141 by an adminUpdates and licensing
Never destructive
Signed releases
Every update is signature-checked before it installs.
Automatic rollback
A backup is taken first, and a failed update restores it.
Fails safe
An expired license stops new work at the next ticket boundary. It never deletes data or touches your repositories.
No third-party code
The dashboard loads no external scripts, fonts or trackers.
Found a vulnerability?
Email [email protected] with the details, and we’ll work with you on a fix and on disclosure.
Bring your security questions
We’ll walk your security team through the architecture, the data flows and the isolation model.