On October 1, 2026, Tuskira released the Tuskira AI Agent Gateway, an open-source, self-hosted project that gives developers, platform teams, and security teams a single control surface between their AI agents and the models and tools those agents use. The gateway runs entirely inside an organization's own environment. No Tuskira account is required, no hosted service is involved, and the organization retains full control of its configuration, policies, and logs.
The Problem: Agent Sprawl Without Visibility
Teams are adopting AI agents faster than they can manage them. Coding assistants such as Claude Code, Cursor, and Codex, AI productivity assistants such as Claude Desktop, and custom agents built in-house each connect to models and tools using their own settings. The result is that organizations often lack a consistent view of what their agents do, what they access, and what they cost.
Model routing, MCP access control, activity logging, and token management are increasingly sold as separate commercial services. Tuskira is bringing them together in one open-source gateway that organizations can operate inside their own boundaries.
"AI agents are moving into production faster than organizations can manage them," said Piyush Sharma, CEO and Co-founder of Tuskira. "Teams are wiring agents to models and tools independently, without a shared view of what those agents do or what they cost. Visibility and control over agent activity should be basic infrastructure, so we're making it open source for any organization to run."
How the Gateway Works
The Tuskira AI Agent Gateway sits in the path between supported AI agents and the LLMs, MCP servers, and tools they use. Supported environments include:
- Coding assistants: Claude Code, Cursor, Codex
- AI productivity assistants: Claude Desktop
- Custom agents, including sub-agents and agent-to-agent workflows
- Automated agents running through pipelines, schedulers, and background jobs
Teams configure models, tools, and access through the gateway instead of managing them separately within each agent. Because the gateway evaluates supported actions before execution, teams can apply policy to both individual actions and which agents can access specific tools.
Core Capabilities
| Capability | What It Does |
|---|---|
| LLM call routing | Send calls to Anthropic, OpenAI, Gemini, and AWS Bedrock through one endpoint with your own keys |
| MCP tool access control | Use centralized profiles to manage which MCP servers and tools each agent may use |
| Per-call enforcement | Each tool call is checked against the agent's profile at execution time. Ungranted calls are denied and every call is logged |
| Agent activity visibility | Record supported model and tool calls, the systems agents contact, and the actions they attempt |
| Token spend tracking | Measure token usage across agents and models to understand adoption and operating cost |
| Local record retention | Store actions, policy decisions, and outcomes inside the organization's environment. Export records to existing logging and analysis tools |
Why Open Source Matters Here
The decision to release this as open source is deliberate. Model routing, access control, and token governance are functions that organizations increasingly need, but the market has been fragmenting them into separate commercial products. By making the gateway open-source and self-hosted, Tuskira ensures that:
- No vendor lock-in. Organizations deploy inside their own boundaries and can inspect, extend, and contribute to the codebase.
- No data leaves the organization. All configuration, policies, and logs remain local. There is no Tuskira-hosted service in the path.
- Community-driven evolution. The project includes contribution guidelines, and the code is available on GitHub at github.com/Tuskira/tusk-ai-secured-gateway.
The release includes installation and deployment guides, documentation for supported agents, LLM providers, and MCP servers, example configurations and access profiles, and policy, logging, and token-tracking guides.
The MCP Governance Gap
The gateway's MCP access control deserves particular attention. The Model Context Protocol has become the de facto standard for connecting AI agents to external tools, but MCP itself does not include a governance layer. Each agent that connects to an MCP server does so with its own permissions, and the organization has no centralized view of which tools are being called, by which agents, and with what frequency.
Tuskira's gateway inserts governance directly into the MCP access path. Instead of managing MCP permissions per-agent, teams create centralized profiles that define which MCP servers and tools each agent may use. The gateway enforces these profiles at execution time, denying ungranted calls and logging every interaction.
This is the kind of infrastructure layer that becomes essential once an organization has more than a handful of agents in production. Uber's MCP gateway, deployed at production scale with 800 servers and 5,000 tools, demonstrated the need at massive scale. Tuskira's open-source approach brings similar controls to organizations that need governance without building a custom platform.
The Broader Agent Governance Landscape
Tuskira is not alone in recognizing the governance gap. Cohere launched North 2 on October 5, 2026, with token spending caps and lockdown mode for enterprise agents. Dataiku released agent management capabilities for cross-platform governance. Oracle introduced Fusion Claw as a governed runtime for autonomous execution. The pattern across all of these announcements is consistent: as agents move from pilots to production, the control plane becomes the critical infrastructure layer.
What distinguishes Tuskira's approach is the open-source, self-hosted model. Enterprise platforms like Cohere North 2 and Oracle Fusion Claw are comprehensive but proprietary. Tuskira's gateway gives organizations that want to maintain full control over their agent infrastructure a viable alternative, with the ability to extend and customize without depending on a vendor's roadmap.
