Technology
Hacker News

Show HN: Coral – Open-source gateway where agents write their own tools

Source Entity

Hacker News

November 17, 2026
Show HN: Coral – Open-source gateway where agents write their own tools

Coral is a new open-source gateway that enhances agent autonomy by allowing AI models to write and reuse their own TypeScript-based tools. The platform prioritizes security by enforcing policy checks and human-in-the-loop approvals before executing sensitive actions.

The Evolution of Autonomous Agent Orchestration

The introduction of Coral marks a significant shift in how developers manage autonomous AI agents. By functioning as a self-hosted gateway, Coral addresses the critical 'execution gap' that exists between an LLM’s reasoning capabilities and its ability to interact with external enterprise systems. Unlike traditional rigid API integrations, Coral empowers agents to write their own TypeScript programs to perform tasks, effectively turning the agent into a developer of its own workflows.

Bridging Reasoning and Execution

Historically, AI agents have been limited by the static tool definitions provided by their developers. Coral breaks this mold by enabling agents to dynamically discover, write, and refine their own tools. When an agent identifies a recurring task—such as cross-referencing a GitHub pull request with a Slack notification—it can codify the interaction logic. This capability reduces the burden on human engineers, who no longer need to pre-build every possible permutation of an agent's functionality, thereby accelerating the deployment of complex automation pipelines.

Security-First Design Principles

The core value proposition of Coral lies in its 'execution rails.' In an era where AI agents are increasingly granted access to sensitive credentials and private data, the risk of unauthorized or harmful actions is high. Coral mitigates this by acting as a policy enforcement layer. Every tool call generated by an agent is intercepted, audited against user-defined security policies, and verified before any credentials are injected. This architecture ensures that even if an agent hallucinates or attempts an unapproved action, the system provides a robust defensive perimeter.

Human-in-the-Loop Integration

Coral’s implementation of a human-in-the-loop (HITL) model is essential for enterprise trust. By requiring manual approval for novel or high-stakes actions—such as the initial posting of a pull request to a messaging platform—Coral creates a feedback loop that validates agent behavior. Once a human user approves a specific code-based tool, the agent can store it for future execution. This transforms the agent from a volatile, experimental entity into a reliable, predictable worker that improves its efficiency over time.

Future Implications for AI Development

As organizations look to scale their AI operations, the ability to govern agent actions without stifling their autonomy will become a competitive advantage. Coral represents a shift toward 'agentic infrastructure,' where the focus moves from simply prompting models to managing the systems they inhabit. By using TypeScript as the medium for tool creation, Coral aligns with standard web development practices, making it accessible to a wide range of developers while maintaining the flexibility required for advanced automation.

Conclusion

In summary, Coral provides a necessary layer of governance for the next generation of autonomous software. By combining creative tool generation with rigorous policy enforcement, it offers a blueprint for how developers can safely harness the power of agentic workflows. As these agents become more sophisticated, tools like Coral will be essential in ensuring that the benefits of autonomous execution do not come at the cost of system security or operational oversight.

Verification Required?

Read the full report from the primary source

Go to Hacker News