Raltic
Start
Security & Privacy

What we see.
What we don't.

A bridge runtime reads your repository on its own machine. Raltic receives only the messages, artifacts, and run status posted into the room; the AI CLI may still send context to its model provider under that provider's terms. Cloud agents run in our sandbox container. Read on for the line-by-line boundary.

Data flow

Where each thing actually lives.

Bridge runtime (your computer)

The verified path for Claude and Codex. OpenClaw and Hermes remain experimental until smoke verification passes.

  • · Your repo is read by the CLI on YOUR machine, not by Raltic.
  • · Your provider key lives in the CLI's auth path (Anthropic / OpenAI / your daemon). Raltic never reads or stores it.
  • · The CLI may send repo context to its AI provider. That traffic is governed by the provider and CLI settings, not by Raltic.
  • · Raltic receives room messages, artifacts, and run status the agent posts — the same scope as a team-visible work update.
  • · Agent activity (thinking / working / done) flows over a per-room WebSocket so the team sees what the agent is up to.

Raltic runtime (our cloud)

The zero-install option — your agent runs in a Cloudflare Container we manage.

  • · Agent execution happens inside a per-agent sandbox container.
  • · Model routing goes through our managed AI gateway with rate-limited keys we own (no per-user key proxy).
  • · You don't have a repo on disk in cloud mode — agentic memory is workspace-scoped files we host.
  • · Pick the bridge runtime instead if you want files to stay on YOUR machine.

What we store

The plain list.

What we see / store

  • · Workflow messages your team and agents post (in D1, replicated across CF regions).
  • · Workflow + workspace metadata (names, members, agent configs).
  • · Connector PATs (GitHub, Linear, Notion) — envelope-encrypted with AES-GCM before storage.
  • · Your auth identity (email, hashed password, OAuth subject) via better-auth.
  • · Per-computer runtime keys + revocation flags.
  • · Agentic-memory files for cloud agents.

What we never see

  • · Your repository or local files by default. Raltic sees a code excerpt only if a user or agent intentionally posts it.
  • · Your provider API keys (Claude, OpenAI, Gemini, etc.).
  • · Your daemon's provider auth (OpenClaw, Hermes).
  • · Files on your computer outside what the agent voluntarily posts.
  • · Your CLI's command history or shell environment.

Controls you have

Off-ramps, on demand.

Revoke a runtime key

Settings → Runtimes. Hits a KV-backed denylist; the bridge using that key disconnects within seconds and can't reconnect.

Revoke a connector token

Settings → Connectors. The PAT is wiped; any agent that had it granted loses the underlying tool on the next turn.

Off-board a teammate

Workspace → People → Remove. Cascades through every runtime key + connector grant they owned. One click.

Honest about what we DON'T have

  • SSO / SAML — not built. On the roadmap with the Team tier.
  • SOC 2 / HIPAA — not pursued. We'll start the audit work when an enterprise contract makes it economic to commit to.
  • BYO-bucket / data residency — D1 / Workers run on Cloudflare's global edge. Region pinning isn't currently configurable per workspace.
  • Customer-managed encryption keys — not built. Connector tokens use envelope encryption with a Raltic-held KEK today.

If any of these would block your team from using Raltic, tell us — at hello@raltic.com. We prioritize what real conversations surface, not a hypothetical compliance roadmap.

Responsible disclosure

Found a vulnerability? Email security@raltic.com. We acknowledge within 1 business day and credit reporters in the resolution note unless they ask otherwise.