kheish  by graniet

AI agent control plane for persistent, tool-using workflows

Created 1 year ago
304 stars

Top 87.8% on SourcePulse

GitHubView on GitHub
Project Summary

Summary

Kheish provides a persistent runtime and control plane for long-lived, tool-using AI agents, ensuring their execution is durable, scoped, and accountable independently of any caller. It targets developers building complex AI workflows that must survive interruptions like process restarts or network outages, offering a robust alternative to in-process agent models.

How It Works

The core invariant is "agents outlive callers." Kheish decouples agent execution from its initiator (CLI, Slack, webhook, code) by treating agents as daemon-managed resources. This daemon enforces durable, journaled execution that survives interruptions like machine reboots or network blips, resuming seamlessly. Actions are governed under named authorities with append-only audit trails, utilizing brokered, short-lived leases for credentials rather than long-lived secrets, enhancing security and accountability.

Quick Start & Requirements

  • Install/Run: Build the daemon binary using cargo build -p kheish-daemon. Start the daemon with export PROVIDER_API_KEY=... (e.g., ANTHROPIC_API_KEY) and ./target/debug/kheish-daemon serve. Interact via client commands (status, sessions create, sessions input, etc.) in a separate terminal.
  • Prerequisites: Rust toolchain, AI provider API keys (Anthropic, OpenAI, Google, xAI).
  • Links: docs/capture.md, AGENTS.md, LICENSE.

Highlighted Details

  • Durable Execution: Agents persist state and resume automatically after interruptions (process restarts, reboots, network failures).
  • Caller/Execution Separation: Agent runs are independent of the caller's lifecycle.
  • Governed Actions: All agent actions are auditable, with support for approvals, leases, and signed external-action records.
  • Brokered Runtime Auth: Manages credentials via revocable, short-lived leases per session/agent.
  • Durable Observations: Ingests and materializes external context (screenshots, webcam, etc.) as durable, policy-governed observations for agents.
  • Multi-Provider Support: Integrates with Anthropic, OpenAI, Google, and xAI.

Maintenance & Community

No specific details on contributors, sponsorships, or community channels were found in the provided README text.

Licensing & Compatibility

The license is available at LICENSE. Specific license type and compatibility notes for commercial use are not detailed in the provided text.

Limitations & Caveats

Kheish is not a coding agent, LLM router, or chatbot framework itself, though it can host these. It is not intended for simple, single LLM calls; direct SDK usage is recommended for such cases. Security requires explicit configuration before exposing the daemon beyond localhost.

Health Check
Last Commit

1 week ago

Responsiveness

Inactive

Pull Requests (30d)
0
Issues (30d)
0
Star History
162 stars in the last 30 days

Explore Similar Projects

Starred by Lilian Weng Lilian Weng(Cofounder of Thinking Machines Lab), Chip Huyen Chip Huyen(Author of "AI Engineering", "Designing Machine Learning Systems"), and
59 more.

AutoGPT by Significant-Gravitas

0.1%
185k
AI agent platform for building, deploying, and running autonomous workflows
Created 3 years ago
Updated 19 hours ago
Feedback? Help us improve.