Repository navigation
Current v0.1.0 — Initial Release
Current is an experimental, open-source runtime for controlled, auditable AI agent execution.
Agents produce typed proposals. Current validates them, requires an explicit approval and separate application step, and records the workflow in a hash-chained commit log.
The agent proposes. The runtime acts. The record is the source of truth.
What's included
- Typed intents for file operations, commands, observations, claims, and reports.
- Explicit proposal review, approval, rejection, and revision.
- A single-node, event-sourced runtime with a hash-chained write-ahead log.
- Durable workflow history, recovery, and content-addressed artifacts.
- OpenAI and Anthropic model integration.
- CLI, HTTP API, and MCP interfaces.
- Default-deny command execution allowlist.
Status
This is an early experimental release, designed primarily for local, single-user development.
Current is not a sandbox or an authentication system. Approved commands execute with the privileges available to the runtime. See the README and threat model for details.
Getting started
See the [README](https://github.com/santosh79/current#readme) for installation, examples, and architecture documentation.