openinterpreter/openinterpreter
A coding agent for open models like Kimi K3
How openinterpreter is put together
This repository is the OpenAI Codex monorepo (branded 'codex'), a locally-running coding agent CLI. The bulk of the system is a large Rust Cargo workspace (`codex-rs`) composed of ~150 crates covering an app-server (JSON-RPC), an Agent Client Protocol (ACP) server, sandboxing (Linux bubblewrap, seccomp/AppArmor), authentication/login, MCP client/server integration, a TUI, and execution/approval policy engines. A thin Node/`bin` CLI shim (`codex-cli`) and a TypeScript SDK (`sdk/typescript`) wrap the Rust core, while Python tooling supports linting. Data flows from the CLI/TUI or ACP client through an in-process app-server client that dispatches typed JSON-RPC requests to the embedded core runtime, which executes sandboxed commands and streams events back. Builds are driven by both Cargo and Bazel, with Docker/devcontainer and Nix profiles for reproducible, network-restricted environments.
Languages
Frameworks
Datastores
Infrastructure
Major components
codex-rs Cargo workspace
Rust monorepo of ~150 crates implementing the core agent, protocols, sandboxing, and tooling.
app-server / app-server-client
Embedded JSON-RPC app server plus an in-process client facade that dispatches typed requests and streams server events to CLI surfaces (TUI, exec).
acp-server
Implements the Agent Client Protocol agent, exposing sessions, auth methods, prompts, and tool-call updates over stdio.
Sandboxing subsystem
Linux bubblewrap/seccomp sandbox (bwrap, linux-sandbox, sandboxing, process-hardening) plus a Windows sandbox crate for isolating executed commands.
Auth & login (login, aws-auth, keyring-store, secrets)
Handles API-key/OpenAI/ChatGPT authentication, workload/agent identity, and credential storage via the OS keyring.
Execution & policy (exec, exec-server, execpolicy, shell-command, shell-escalation, tools)
Runs shell/tool commands under approval and exec-policy constraints with escalation controls.
codex-cli & TypeScript SDK
Node bin shim (`bin/codex.js`) and an ESM TypeScript SDK packaging the Codex CLI for distribution.
Dev container / secure runtime profiles
Contributor and hardened 'secure' devcontainers with firewall allowlists, IPv6 blocking, and bubblewrap-in-Docker capability tuning.
Shipped the 0.0.40 release and cleaned up compatibility and build problems from syncing with the upstream Codex 0.148.0 engine.
3 weeksWeek of 2026-08-17
Week of 2026-08-17latest
Shipped the 0.0.40 release and cleaned up compatibility and build problems from syncing with the upstream Codex 0.148.0 engine.
Week of 2026-08-10
A very busy week of rebranding to Open Interpreter, tightening permissions and auditing, and cutting three releases (0.0.37–0.0.38).
Week of 2026-08-03
Built the foundation of the new remote 'code-mode' execution service along with a hooks framework and stronger review and sandboxing controls.
Over the last few weeks, Open Interpreter has been rebranding and re-basing its command-line tool on the upstream Codex engine, cutting a rapid string of releases (0.0.37 through 0.0.40) while fixing the compatibility and build issues that came with each sync. Alongside the version churn, the team built out a new 'code-mode' remote execution system, a plugin/hook framework, and stronger permission and auditing controls. The most recent week narrowed to release cleanup and stabilization after the largest burst of feature work.
Week by week
2026-08-17Shipped the 0.0.40 release and cleaned up compatibility and build problems from syncing with the upstream Codex 0.148.0 engine.latest6 changes
Chore
Released version 0.0.40
Prepared and merged the 0.0.40 release of Open Interpreter.
Chore
Synced with upstream Codex 0.148.0
Brought the tool in line with the latest version of the underlying Codex engine it is built on.
Fix
Fixed multiple build and integration failures
Resolved a batch of broken builds and test failures that surfaced after the engine sync.
Fix
Fixed a sign-in hang
Stopped the model picker from hanging on a stale saved login.
Fix
Fixed continuous-integration setup
Corrected the Ubuntu software-source configuration used by the automated build system.
Chore
Restored the Rust release workflow
Brought back the automated pipeline that packages and publishes the tool.
2026-08-10A very busy week of rebranding to Open Interpreter, tightening permissions and auditing, and cutting three releases (0.0.37–0.0.38).6 changes
Chore
Released versions 0.0.37 and 0.0.38
Prepared and shipped two point releases during the week.
Refactor
Centralized Open Interpreter branding
Consolidated the product name and identity across the interface, error messages, and documentation.
Feature
Added plugin and hook support
Introduced a system that lets external tools plug in and refresh automatically when plugins change.
Feature
Added per-environment permission controls
Made it possible to enforce different approval and access rules depending on the working context.
Feature
Added storage and startup diagnostics
Gave the built-in health checker insight into storage and showed resume status during startup.
Feature
Added a health check endpoint for remote execution
Exposed a way to confirm the new code-execution service is running.
2026-08-03Built the foundation of the new remote 'code-mode' execution service along with a hooks framework and stronger review and sandboxing controls.6 changes
Feature
Built the code-mode host service
Created the core service and network protocol for running code in a managed remote host.
Feature
Added command hooks
Let users attach custom actions that run before or after commands, including asynchronous ones.
Feature
Added automatic review for managed models
Enforced approval reviews for certain models to keep risky commands in check.
Feature
Added identity token exchange
Introduced secure identity-based authentication for connecting to services.
Refactor
Improved responsiveness of live connections
Tuned network settings and kept sessions alive through connection drops for smoother real-time use.
Refactor
Reorganized the skills system
Removed legacy skill components and moved skill handling into a cleaner extension.
Dependencies and code review
Dependency advisories
Security Watch
No known advisories across 1 scanned dependencies.
No known advisories in the scanned dependencies.
Code review
No concrete code-level issues confirmed.
Get this report every week for your repos.
GitZoid learns each repo, reports what changed, and flags what needs attention. One flat price for the whole team.
$19 a month, flat · First 10 outputs free · No card required