portunus
mdostal/portunusBoundary-only secret broker keeping credentials out of LLM contexts
PersonalEnterpriseAI & LLM toolingMCP serversSecurity & identityDocker / self-hostCLIPermissiveAt a glance
- stars on GitHub
- 2
- contributors
- 3+
- last commit
- latest release, 2 weeks ago
- v0.34.0
- license
- MIT
About
Portunus is a standalone secret broker that allows AI agents to use vault secrets without exposing raw credentials to the LLM's context, logs, or tool outputs. It resolves secrets using metadata tags and injects values strictly at the execution boundary via subprocesses, environment variables, or temporary files. It is run primarily through a command-line interface, as a Model Context Protocol (MCP) server, or in Docker containers.
From the repository: “A standalone, boundary-only secret broker — reference a secret by name, the value is injected only at the execution boundary, never inside an LLM or agent context. Free and open source.”
Downloads
All releases on GitHub ↗Latest release: v0.34.0 — Chrome saved-login backend, . Its files were downloaded 4 times.
All files (1)
Other files
- portunus-desktop-0.34.0.zip21 MB · 4 downloads
These files come from the project's GitHub release. DigGitHub does not check them.
Alternatives
All alternatives →| Project | Stars | Last commit | License | Worth |
|---|---|---|---|---|
 portunusThis project | 2 | MIT | 53 (best) | |
 agent-vaultLocal credential proxy for coding agents | 0 | Apache-2.0 | 37 | |
 keywardLocal, encrypted secret broker for AI coding agents | 1 | MIT | 33 | |
 mcp-wardenLocal-first credential broker and MCP server for AI agents | 9 | Apache-2.0 | 50 | |
 agentgateHuman-in-the-loop approval workflows for AI agents | 33 (best) | MIT | 50 |
▲ marks the most stars and the best worth score.