codeform.io

What It Is

Codeform is an AI coding harness aimed at “serious coding work.” Its official site explicitly says it is hard-forked from opencode, and that it is proprietary rather than open source. It is not a standalone large-model product; instead, it adds a layer of engineering discipline between models such as Anthropic, OpenAI, Google, Groq, local Ollama/llama.cpp, and your codebase. The goal is to move from a “diff that looks correct” to a “PR that actually works.”

Core Capabilities

Its main highlights are Agent Discipline, Symbolic Intelligence, Plans, and Team Mode. Agent Supervision can detect repeated file reads, tool-call loops, long periods without edits, empty outputs, repeated text, invalid tool parameters, and more, intervening at passive, active, or aggressive levels. Symbolic Intelligence uses LSP to understand functions, classes, types, and references, supporting symbol_overview, symbol_find, symbol_references, and symbol_edit. This makes it better suited to large codebases than full-text grep and text replacement. Plans are saved to disk as Markdown and can be resumed across sessions. Team Mode lets a lead coordinate multiple members working in parallel in separate git worktrees, with shared tasks, mailboxes, recovery, and audit logs.

Pricing and Access

The official site does not disclose product subscription pricing, free quotas, or a public trial. It only states that Codeform is currently available via private access, and that users need to request a download link by email. Users must bring their own LLM provider API key, so actual model costs depend on external vendors. The documentation notes that Team Mode has a default hard cost cap of $50 per run, and supports pre-run cost estimation as well as runtime cost limits.

Pros and Cons

Its strengths lie in its fairly complete engineering-governance capabilities: models can be swapped, provider failures can fall back to alternatives, long-running tasks can use context recovery, and model drift can be monitored. For teams, plans, templates, multi-agent workflows, audit logs, and cost controls are all practical features. The drawbacks are also clear: private access makes purchasing, deployment, and support information opaque; the configuration surface is broad, creating a higher learning curve; Windows users are advised to use WSL, suggesting the native experience may be limited; and Chinese-language support, privacy compliance, and enterprise SLA details are not stated in the main content.

Who It’s For and Access from China

Codeform is better suited to professional development teams with complex existing codebases, teams willing to bring their own model APIs, and those that prioritize controllability and recoverability. It is less suitable for beginners who only want lightweight code completion. There is no information in the main content about access from mainland China, so its availability there is unknown; payment methods are also not disclosed. Alternatives to compare include Claude Code, Cursor, Windsurf, and opencode.