Variant Technologies
Variant Technologies presents itself as the developer of Benzi, a compiler-backed AI coding agent. Benzi compiles a codebase into a resolved, queryable map of calls, data flow, control flow and class hierarchy, then uses that map to investigate bugs, explain and modify code, run tests and verify repairs while keeping source local and letting users bring their own model/API key. (Sources: https://varianttech.net/, https://github.com/oooscoos/Benzi)
At a Glance
- Individual developers
- Software engineering teams working in large or unfamiliar repositories
- AI coding-agent users who need deterministic code intelligence
- Teams using Claude Code, Cursor or other MCP clients
- +1 more
AI Tools by Variant Technologies
(1)Benzi
Compiler First AI Coding Agent
Discussions
No discussions yet
Be the first to start a discussion about Variant Technologies
Latest News
Benzi repository adds inline intro video and publishes benchmark traces comparing Claude Code and Benzi
Second SWE-bench run reports $9.26 total cost for 500 tasks, or 1.85 cents per task
October SWE-bench submission kit reports 412/500 (82.4%) on DeepSeek V4.1-Flash
Official SWE-bench regrade reports 391/500 (78.2%) and 9.5 cents per resolved instance
Products & Services
Compiler-backed autonomous coding agent and code-intelligence infrastructure. It builds a resolved map of calls, data flow, control flow and class hierarchy before answering questions or making changes; it supports local compilation, bring-your-own-model, runtime tracing, persistent repository memory, parse-gated edits, receipts and rollback. It is offered free during beta. (Sources: https://varianttech.net/, https://github.com/oooscoos/Benzi, https://pypi.org/project/benzi/)
The full Benzi agent delivered through a VS Code extension, with a settings/login panel and the agent loop operating on the compiled index. (Sources: https://github.com/oooscoos/Benzi, https://marketplace.visualstudio.com/items?itemName=varianttech.benzi)
Terminal/CI form of the Benzi agent. The package installs benzi-login, benzi-headless and benzi-mcp commands and supports Windows, macOS and Linux with Python 3.11–3.14. (Source: https://github.com/oooscoos/Benzi)
The compiler/index served as MCP tools for existing agents such as Claude Code, Cursor or another MCP client, without requiring Benzi’s own agent loop. (Source: https://github.com/oooscoos/Benzi)
Market Position
Variant positions Benzi against general-purpose coding agents and code indexers by compiling the repository first rather than relying primarily on grep/glob or embeddings. Its differentiators are compiler-resolved structural relationships, explicit edge-confidence labels, local processing, runtime confirmation, model agnosticism and MCP delivery. The README explicitly contrasts it with Claude Code, Cursor and Aider; relevant adjacent alternatives include those tools and conventional LSP/SCIP indexers. (Source: https://github.com/oooscoos/Benzi)
Leadership
Founders
Shobhit Srivastava
LinkedIn identifies Shobhit Srivastava as “building benzi,” with education at the University of Washington and prior experience at AWS; his public profile/search result describes a broad software background spanning game development, deep-learning model development/deployment, cloud infrastructure and app development. (Source: https://www.linkedin.com/in/shobhit858)
Founding Story
The public materials frame Variant Technologies as starting with Benzi to address limitations the creator sees in AI coding workflows: conventional agents search or read raw source, whereas Benzi first compiles the repository into a deterministic structural map. The GitHub profile says, “Starting with Benzi, we aim to solve major issues we see in AI industry” and emphasizes that users’ data remains theirs. (Sources: https://github.com/oooscoos, https://github.com/oooscoos/Benzi)
Business Model
Revenue Model
The public materials describe a beta product that is free to use. Users supply their own model/API key; the compiler runs locally. No paid commercial pricing or usage-billing model was identified in the reviewed sources.
Pricing Tiers
Benzi is stated to be completely free during beta; available as the VS Code agent, headless CLI and MCP compiler server. Users bring their own model/API key. (Source: https://github.com/oooscoos/Benzi)
Target Markets
- Individual developers
- Software engineering teams working in large or unfamiliar repositories
- AI coding-agent users who need deterministic code intelligence
- Teams using Claude Code, Cursor or other MCP clients
- Developers using VS Code, terminal, scripts or CI
- Autonomous bug investigation and repair
- Building features across an entire repository
- Explaining unfamiliar codebases and tracing callers/callees
- Tracing data origins and return-value destinations
- Running tests and repairing failures
- Greenfield application generation, demonstrated by the StallionSwipe horse-dating-app example