Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Orchestration engine for long-running, tool-using AI agents
| Date | Stars |
|---|---|
| 2026-07-31 | 262 |
| 2026-08-06 | 262 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
15.0
growth rate 0.00%/day
<p align="center"> <img src="docs/logo.png" alt="Kheish" width="180"> </p> <h1 align="center">Kheish</h1> <p align="center"> <em>Persistent agent runtime and control plane for long-lived, tool-using AI workflows.</em> </p> <p align="center"> <a href="https://kheish.ai"><img src="https://img.shields.io/badge/Website-kheish.ai-ED1C2E?style=for-the-badge&logo=firefoxbrowser&logoColor=white" alt="Website"></a> <a href="https://docs.kheish.ai"><img src="https://img.shields.io/badge/Docs-docs.kheish.ai-0B0C0E?style=for-the-badge&logo=readthedocs&logoColor=white" alt="Documentation"></a> </p> --- ## The invariant: agents outlive callers Kheish enforces **one rule**: > An agent's execution is durable, scoped, and accountable — independently of any caller, process, or human reachability. Everything else in this README is a consequence of that single invariant. If a feature does not protect this rule, it does not belong in the daemon. The rule has three properties, and they are not separable: - **Caller / execution separation.** The thing that *asks* for an agent run is not the thing that *runs* it. A CLI, a Slack message, a webhook, your code — all of them submit; the daemon executes. The caller can crash, disconnect, log out, get rate-limited, or never come back. The run keeps going. - **Survives interruptions.** Process restart, machine reboot, network blip, missing human, expired credential, paused approval, deferred question — none of these are exceptions. They are normal points in the run's lifecycle, journaled, replay-safe, and resumed by the daemon when the world is ready again. - **Governed.** Every action an agent takes happens under a named authority (a session, a sidechain, a brokered lease) with a stated scope and an append-only audit trail. Approvals, leases, scope narrowing, and signed external-action records are not optional features — they are how the rule is upheld. That is the product. The daemon, the connectors, the capture pipeline, the multi-provider routing, the skills, the schedules — they exist to keep that one rule true under real-world conditions. ## What this addresses LangChain, AutoGen, CrewAI and vanilla SDK loops are excellent at composing agents *inside an application*. Kheish addresses a different failure mode: **what happens when the application, the human, or the credential boundary disappears while the run is still alive.** In the in-process model, the agent and the caller share a fate: - the agent lives inside the caller — caller dies, run dies - state is in process memory — reboot is data loss - the agent runs with the caller's credentials — every script holds long-lived secrets - "human in the loop" is a blocked thread waiting on stdin - audit is whatever you remembered to log Kheish decouples them. The agent is a daemon-managed resource, like a database row or a workflow execution. Callers — a CLI, Slack, a webhook, your code — submit and observe; they do not host the run. Composition stays in your application; execution moves into the daemon. ## What Kheish is not - Not a **coding agent** like OpenHands / Aider / Cursor. You can run one on top of Kheish; coding is one workload among many. - Not an **LLM router** like LiteLLM. Routing is included, but Kheish is the layer above. - Not a **chatbot framework**. Slack and Telegram are *one* ingress among many. - Not the right tool if you only need a single LLM call. Use the SDK directly. The thing to read this against: **Kubernetes is a control plane for containers. Temporal is a control plane for workflows. Kheish is a control plane for AI agents.** Same shape, same reason it exists — to make the running thing outlive the caller, and to make the operator the source of truth instead of the script. ## Quickstart Build the daemon binary; it acts as the server *and* the CLI client. ```bash cargo build -p kheish-daemon ``` Start the daemon. By default it binds `127.0.0.1:4000` and journals state under `./.kheish-da
Excerpt of 8,749 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:a721f689285b8be2, topic:ai-agent, topic:ai-agents, topic:multi-agent