The Palo Alto company's TokenOps feature alerts teams at 80% of budget and s AI agents when they hit their limit
AI agents are very good at doing work. They are also very good at spending money while nobody is watching.
Nasiko, a Palo Alto-based tech company, wants to fix the second part. On November 16, 2026, it announced TokenOps, a set of spending-tracking and budget-control features built into its OpenRuntime platform. The pitch is simple: see what every agent costs, and stop it automatically when it runs out of allowance.
What TokenOps actually does #
Operators can set a budget for individual agents. When spending reaches 80% of the configured limit, the system sends an alert. When spending hits 100%, it s the agent’s operations automatically.
The feature also breaks down token usage and associated costs in three ways: by individual agent, by team, and by model. Tokens are the units AI providers use to meter and bill usage. More tokens consumed means a larger bill, so tracking them is essentially tracking the meter.
How the platform sits in the middle #
TokenOps is a feature, not a standalone product. It lives inside OpenRuntime, which Nasiko describes as a control plane for AI agents.
Technically, OpenRuntime runs as a single server process that proxies the calls AI agents make. Because every request flows through it, the platform can attribute costs across multiple providers, including OpenAI and Gemini. That cross-provider view is the core of the spending tracker. Without a middle layer, each provider’s bill shows up separately, and stitching them together becomes someone’s spreadsheet hobby.
The open-source runtime bundles several other tools alongside budget controls:
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
- An LLM router for model routing, which directs requests to different models
- Permission management through an MCP gateway, handling which agents can access which tools
- Mechanisms to prevent infinite loops, where an agent keeps calling itself or a tool without end
- Observability through OpenTelemetry traces, including automatic logging of tokens used and costs incurred
On the routing side, potential token spend reductions are estimated at up to 50% through effective routing.
Compatibility without a rewrite #
According to the project’s materials, no SDK migration is necessary. Existing agents and workflows can keep running as they are while teams adopt centralized budget governance.
The platform also supports popular coding agents, including Claude Code and Codex. Custom agents can be deployed in a containerized environment using Docker.
Operators get a choice of deployment models. They can self-host the platform or use a hosted option.
The project is publicly available on GitHub, where it has drawn significant community engagement.
The road to TokenOps #
TokenOps is the latest step in a year of steady releases for Nasiko.
The company introduced its core control plane in May 2026. In September, it onboarded onto the DigitalOcean Marketplace, giving the platform another distribution channel. The November launch layers financial controls on top of the infrastructure that was already in place.
Nasiko frames the broader problem around what it calls the “agent estate.” The term describes the growing collection of AI agents that teams run across different frameworks. That sprawl is the gap Nasiko says it is trying to fill. Individual agent frameworks may track their own usage, but a cross-framework, cross-provider view requires something that sits above all of them.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.