Freedom251 › AI Coding Assistants
Agent Code
What Agent Code is
Agent Code is a free, MIT-licensed AI coding agent for terminal-based software work. It reads codebases, runs commands, edits files, and iterates on tasks. You can use an interactive terminal interface, headless scripts and CI, a local HTTP API, or an Agent Client Protocol bridge for IDEs. The project lists more than 30 tools covering files, shell commands, search, web, LSP, MCP, worktrees, and subagents. It supports model providers including Anthropic, OpenAI, Azure OpenAI, xAI, Google, OpenRouter, and local OpenAI-compatible APIs. Skills, plugins, project conventions, resumable sessions, and a security scanning mode are included. Tool calls pass through permission checks, with ask mode as the default and a read-only plan mode available. Write tools are blocked from .git/, .husky/, and node_modules/. Session data is stored locally, and conversation history leaves the machine only for LLM API calls. Telemetry is opt-in. Installation options include Cargo, Homebrew, Docker, prebuilt binaries, and source builds.
Who it suits
Agent Code suits developers who want an open-source coding agent across terminal, CI, HTTP, or IDE workflows. It requires an LLM API credential or subscription login, and full tooling requires git and ripgrep.
What it does well
- MIT-licensed and free to use
- Supports terminal, CI, HTTP, and IDE modes
- Lists more than 30 development tools
- Permission checks govern tool calls
- Session data is stored locally
What to know first
- Requires an LLM credential or subscription login
- Full tooling requires git and ripgrep
- LLM endpoint security is outside its threat model
- Telemetry requires opting out?
Plans
| Plan | Billed and limits | Price |
|---|---|---|
| Open source | MIT-licensed software; requires a supported LLM credential | Free |
As github.com lists them, read 12 Oct 2026. The maker's own page has the final word.
Verdict
Agent Code offers broad execution modes, model-provider support, and documented safeguards for coding tasks. Its free license is useful for technical users who can provide the required model access and local dependencies.
Facts from Agent Code's own pages
- Coding agent
- Yes · github.com, 22 Sept 2026
- Chat assistance
- Yes · github.com, 22 Sept 2026
- Data handling
- The security policy says session data is stored locally and conversation history leaves the machine only for LLM API calls. · github.com, 12 Oct 2026
- Features
- The project lists 30+ tools for files, shell, search, worktrees, web, LSP, MCP, and subagents, along with skills, plugins, and resumable sessions. · github.com, 12 Oct 2026
- Free plan
- Yes · github.com, 22 Sept 2026
- Install options
- The README lists installation through Cargo, Homebrew, Docker, prebuilt binaries, or building from source. · github.com, 12 Oct 2026
- Integrations
- It supports MCP and ACP integrations and exposes a local HTTP API for clients, including the project’s Flutter GUI. · github.com, 12 Oct 2026
- License
- Agent Code is released under the MIT license. · github.com, 12 Oct 2026
- Maker
- The GitHub organization identifies the maker as Avala AI and lists the United States of America as its location. · github.com, 12 Oct 2026
- Model providers
- It supports providers including Anthropic, OpenAI, Azure OpenAI, xAI, Google, OpenRouter, and local OpenAI-compatible APIs. · github.com, 12 Oct 2026
- Open source
- The project is MIT licensed. · github.com, 12 Oct 2026
- Permissions
- Its security policy says tool calls pass through permission checks, with ask mode as the default and a read-only plan mode. · github.com, 12 Oct 2026
- Platforms
- Prebuilt binaries are listed for macOS, Linux, and Windows. · github.com, 12 Oct 2026
- Protected paths
- Write tools are blocked from modifying `.git/`, `.husky/`, and `node_modules/`. · github.com, 12 Oct 2026
- Purpose
- Agent Code is an open-source AI coding agent for the terminal that reads codebases, runs commands, edits files, and iterates on tasks. · avala-ai.github.io, 12 Oct 2026
- Requirements
- Installation requires an LLM API credential or subscription login; full tooling requires git and ripgrep, and Linux sandboxing optionally uses bwrap. · github.com, 12 Oct 2026
- Security controls
- The project documents permission modes, protected directories, destructive-command checks, system-path protections, and an optional OS sandbox. · github.com, 12 Oct 2026
- Security limits
- The threat model excludes the security of the LLM API endpoint and the user's local machine beyond what Agent Code touches. · github.com, 12 Oct 2026
- Security support
- The security policy asks users to report vulnerabilities by email, promises acknowledgment within 48 hours, and gives a fix-timeline response within five business days. · github.com, 12 Oct 2026
- Support
- The security policy asks vulnerability reports to be emailed to [email protected] and says reports are acknowledged within 48 hours. · github.com, 12 Oct 2026
- Telemetry
- The README says telemetry is not sent unless the user opts in. · github.com, 12 Oct 2026
- Terminal access
- Yes · github.com, 22 Sept 2026
- Tools
- The project lists more than 30 tools, including file operations, shell, search, web, LSP, MCP, and subagents. · github.com, 12 Oct 2026
- Usage modes
- It supports an interactive terminal UI, headless scripts and CI, an HTTP server, and the Agent Client Protocol for IDE integrations. · github.com, 12 Oct 2026
Where Agent Code stands on Freedom251
- AI Coding Assistants#38 of 103