intent-systems.com

What It Is

Intent Systems is not another code generation model or IDE. Instead, it is context infrastructure that builds an “Intent Layer” for complex codebases. It compresses team knowledge—source code, data sources, infrastructure, git history, code reviews, and more—into structured context, so AI coding agents such as Cursor, Claude Code, Copilot, and Codex understand the project before starting a task.

Core Capabilities and Results

Its core value is reducing the cost of getting AI agents up to speed on a codebase from scratch. The company says agents can more accurately find relevant files, dependencies, and cross-module relationships, reducing hallucinated imports and missed dependencies. It also supports complex tasks such as cross-service refactoring, data migrations, and multi-module feature development. Its evaluation approach is fairly practical: it uses real customer backlog tasks, compares outputs with and without the Intent Layer, and scores them by correctness, completeness, and quality. The materials cite research results such as Roblox code acceptance rates rising from 30% to 60%, and a 3.3x accuracy improvement across 300+ open-source repositories. However, they also acknowledge that improvements vary by task, and smaller codebases may see limited benefit.

Pricing and Delivery

Pricing is clearly enterprise-oriented. The Proof Pilot costs $20,000 per 1M tokens of mapped source code, including onboarding and evaluation, with delivery promised within 48 hours. If the customer’s evaluation shows no improvement, there is no charge, and the customer can still keep the Intent Layer. Further expansion is also priced at $20,000 per 1M tokens. Ongoing maintenance, called the Freshness Retainer, costs $5,000–10,000 per month and covers automatic updates after commits/merges, staleness detection, and priority support. Its main free component is the open-source tokcount tool, used to estimate a codebase’s token count.

Privacy, Security, and Integrations

The official security positioning is fairly complete: read-only repository access, standard NDAs, and the option to use the customer’s own API keys and preferred LLM provider. Code processing can go through SOC2-certified Anthropic or OpenAI, and API data is not used for training. For security-sensitive teams, the cartographer can also run inside the customer’s own infrastructure, preventing code from leaving the network. It does not require replacing existing development tools; instead, it acts as a context layer that enhances current AI coding agents.

Pros, Cons, and Best Fit

Its strengths include before-and-after evaluation on real code, compatibility with existing tools, a strong focus on secure deployment, and a no-improvement-no-fee mechanism. The downsides are high pricing, the lack of a self-service SaaS model, and the need to provide repository access and cooperate on evaluation tasks. Chinese-language support, payment methods, and access from mainland China are not disclosed. It is best suited for enterprise engineering teams with codebases at the million-token scale or above, whose AI coding agents have hit effectiveness bottlenecks and who are willing to invest budget into improving engineering productivity.

China Access and Alternatives

Based on the available text, it is not possible to determine whether the service is directly accessible from mainland China. There is also no mention of RMB payments, Chinese-language customer support, or local compliance documentation. If networking, model API access, or cross-border code security becomes a barrier, alternatives or complementary options to consider include Sourcegraph Cody, Cursor, Claude Code, GitHub Copilot, Qodo, Continue, and Codeium.