Conclusion

Let’s get straight to the point: the parts for a ‘Zero Yen Monthly AI Environment’ have come together in just one week.

There was a time when paying several hundred dollars a month for cloud API services to run AI agents was considered ‘normal.’ That was just a month ago.

However, in the past week, the following four components have emerged simultaneously:

• CLI ‘8v’ that reduces token usage by 66%
• Local AI memory runtime ‘Squish’
• Open-source coding agent ‘Frontman’ that operates entirely within the browser
• ‘Polynya’ that transforms Postgres into an AI workspace

Each of these is impressive on its own, but when combined, they change the landscape. It has become feasible to operate AI agents at a practical level without monthly payments for cloud APIs.

The question is simple: “Will you still continue to pay for cloud services every month?”

—

66% Reduction in Tokens—How ‘8v’ Changes Cost Structure

The majority of the running costs for AI agents are determined by token consumption. If you hit a model of GPT-4o class via API, it costs about $5 for 1 million input tokens and about $15 for output. When agents autonomously handle tasks, they can easily consume tens of thousands to hundreds of thousands of tokens in a single day. When calculated monthly, even small tasks can lead to costs of several hundred dollars.

‘8v’ is a CLI tool that optimizes interactions between the agent and the model, completing the same tasks with only 34% of the token usage compared to traditional methods. It automatically compresses context, trims unnecessary interactions, and structures prompts.

Let’s think in numbers. Suppose you had a process that incurred $300 in API costs per month. By simply integrating ‘8v’, you could reduce this to about $100 while maintaining the same output quality. That’s a difference of $2,400 annually, or about 360,000 yen. For small and medium-sized enterprises in rural areas, this difference could be a decisive factor between proceeding or not.

Moreover, this calculation assumes you continue using cloud APIs. When combined with the local execution environment mentioned later, costs could drop even further.

—

Local AI Memory ‘Squish’—The End of Relying on Cloud for Memory

For AI agents to behave intelligently, they require ‘memory.’ This includes past conversations, user preferences, and contextual information about business operations. Until now, this memory has mostly been stored in cloud-based vector databases or external APIs, with services like Pinecone and Weaviate requiring monthly payments to store data.

‘Squish’ is a memory runtime that allows this memory to be handled entirely on local machines. It enables agents to retain, search, and update the contextual data they need on their own machines.

This has two significant implications:

The first is cost. Monthly payments for cloud vector databases (even small-scale ones typically range from $20 to $70) can be eliminated.

The second is data sovereignty. There’s no need to upload customer data or internal know-how to the cloud. For small and medium-sized enterprises in rural areas, the sentiment of “we don’t want to expose our data” is not just an emotional stance but a rational risk management strategy. It’s not uncommon for contracts with business partners to prohibit cloud storage. With Squish, data never leaves the company’s machines.

Technically, there is still a limitation that it is “not suitable for large-scale data.” However, for companies with fewer than 50 employees, the amount of data they handle is typically manageable locally. In fact, it’s likely that many cases will find local solutions faster for their scale.

—

Browser-Based Coding Agent ‘Frontman’—No Need for Cloud in Development Environments

When it comes to coding assistance using AI, GitHub Copilot and Cursor are the go-to options. Both are excellent, but they come with monthly subscriptions ranging from $10 to $40. That adds up to an annual cost of 120,000 to 500,000 yen per person. For three developers, that’s 150,000 yen annually.

‘Frontman’ is an open-source coding agent that operates in the browser. By connecting it to local models (like those via Ollama with Llama), you can utilize coding assistance with zero API charges and subscription costs.

Of course, compared to GPT-4o or Claude 4, the output quality of local models may be lower. However, the question to ask here is not “Is the highest quality necessary?” but rather “Is this sufficient for the task at hand?”

Routine CRUD operations, refactoring existing code, generating test code—80% of the tasks frequently encountered in small and medium-sized enterprise development environments can be adequately handled by local models. The remaining 20% of more complex tasks can still utilize cloud APIs. It’s not a binary choice between all cloud or all local.

—

‘Polynya’ Turns Postgres into an AI Workspace—Leveraging What You Already Have

Many small and medium-sized enterprises already use PostgreSQL for their backend systems, customer management, and inventory management. It’s a mature technology with ample operational know-how.

‘Polynya’ is a tool that transforms that Postgres into an AI agent workspace. When AI agents require real-time data, it sets up ephemeral data warehouses on Postgres and deletes them once processing is complete. There’s no need to contract for a constantly running data warehouse (like Snowflake or BigQuery).

The minimum monthly fee for Snowflake starts at several hundred dollars. While BigQuery charges are based on usage, frequent queries by agents can lead to high monthly costs. With Polynya, you simply build on top of the already operational Postgres. Additional infrastructure costs are nearly zero.

This isn’t about “buying something new”; it’s about “increasing the value of what you already have.” For small and medium-sized enterprises, this distinction is significant.

—

What Happens When You Combine the Four—Cost Estimation

Let’s calculate specifically. Assume a rural company with 30 employees incorporates AI agents into its operations.

Traditional Cloud-Dependent Structure:

That amounts to approximately 600,000 to 1,800,000 yen annually. That’s a heavy burden for small and medium-sized enterprises.

Local + Open Source Structure (Utilizing Tools Available This Week):

That’s about 45,000 to 120,000 yen annually. The difference can be as much as 1.7 million yen per year. For rural small and medium-sized enterprises, that amount could be the difference between hiring another employee or not.

—

Response to ‘But Isn’t the Transition a Hassle?’

When discussing these topics, the common response is, “What about the implementation costs?” “What about the learning curve?” “Changing our current methods is difficult.”

To be honest, there’s no need to switch everything all at once.

Start by trying out just ‘8v.’ By integrating it into existing API calls, the setup takes just a few minutes. You can verify for yourself whether token consumption really decreases. If it does, then try ‘Squish’ next. Move the memory of one agent to local storage.

Test small and make decisions based on numbers. This is the correct approach for small and medium-sized enterprises when adopting AI. There’s no need to launch a “company-wide implementation project” like large corporations. In fact, the ability for one person to test it in just two hours in the afternoon is the greatest asset for small and medium-sized enterprises. No need for approvals or committees.

—

What This Trend Means—The Day When ‘AI Costs’ Are No Longer a Competitive Variable

The essence of what has happened this week is the fact that the ‘cost of using AI’ is rapidly approaching zero.

Until now, how much a company could afford to pay monthly for AI determined its level of AI utilization. This structure favored large corporations with financial resources.

However, as costs approach zero, the differentiating factor will become the ideas and execution speed regarding ‘what to have AI do.’ Here, small and medium-sized enterprises, which can make decisions quickly and are closer to the field, will have the advantage.

By saving 1.7 million yen in monthly costs, you can invest that amount in “people who think about what to have AI do.” Alternatively, you can create new customer touchpoints with the time saved. What you do with the reduced costs will be the next decisive factor in the competition.

This week, the parts have come together. All that’s left is to take action. Start by opening the GitHub page for ‘8v.’