Security & Privacy

July 7, 2026 · View on GitHub

Privacy model

Callimachus is local-first. Specifically:

  • Your conversation index (index.db) lives only on your machine and is never uploaded anywhere.
  • Embeddings are computed on-device — no text is sent to a cloud embedding API.
  • API keys are stored in the OS keychain, never in plaintext on disk or in the repo.
  • The only outbound network traffic is to the LLM provider you explicitly pick and the one-time embedding-model download on first index. Conversation content leaves your machine only when you use a cloud provider for one of three features: (1) the in-app chat, (2) Knowledge distillation (agent::distill) of decisions, gotchas, and summaries, and (3) Ask your history / RAG (agent::answer). Choosing local Ollama at its default local address keeps distillation and Ask fully on-device. Pointing its custom server URL at a remote host — even one you control (a VPS, a homelab box) — transmits conversation content off the local machine; it stays within your own infrastructure, but it is no longer on-device. Ollama Cloud (ollama.com) is a cloud provider like any other: picking it sends conversation content to Ollama's hosted service, keyed by your Ollama API key. Indexing, search, and embeddings are always local regardless of which engine you pick.
  • The in-app agent's run_shell tool requires explicit per-command approval before anything executes.

Reporting a vulnerability

Please do not open a public issue for security vulnerabilities.

Email ari@shaller.dev with:

  • a description of the issue and its impact,
  • steps to reproduce (a proof-of-concept if possible),
  • any suggested remediation.

You'll get an acknowledgement within a few days. Once a fix is released, we're happy to credit you (or keep you anonymous — your call).

Supported versions

The latest released version receives security fixes. As a pre-1.0 project, fixes ship in the next patch release rather than as backports.