Independent register of AI agents & harnesses · no sponsored placementsEdition 2026-09 · 107 entries · evidence to 26 Sept 2026

The Register / Coding agents / Open Interpreter

Register entry · Coding agents · updated 26 Sept 2026

Open Interpreter

Open Interpreter now ships as a Rust fork of the OpenAI Codex CLI that tracks upstream Codex releases and adds provider-neutral model configuration, Chat Completions and Anthropic Messages transports, and a /harness switch that emulates the agent loops of Claude Code, Kimi Code, Qwen Code, ZCode, DeepSeek TUI and SWE-agent. It keeps Codex's sandbox modes, approval policies, MCP, AGENTS.md, skills, hooks and ACP support, and also has a desktop app. It is Apache-2.0 licensed and free; users pay their model provider.

What reviewers and users praise

  • Brings a mature Codex terminal interface to cheap and open-weight models (project docs; user reports of pairing with DeepSeek)
  • Very frequent releases tracking upstream Codex (20 GitHub releases in 90 days)

What they criticise

  • Provider-compatibility bugs reported on GitHub (tool calls failing on Moonshot K3, model picker hangs, Ollama/Qwen loops)
  • Little independent review or community discussion of the relaunched product so far
Rating certificate · AW/26-09/52226 Sept 2026

Open Interpreter

Open Interpreter (Killian Lucas) · Coding agents

Trust AOpen source
Adoption90
Experts—
Crowd—
Experts vs the crowdneeds both scores
Experts—
Crowd—
PriceFree
Use it viaCLI, Desktop app, IDE extension, MCP server
Runs onmacos, linux, windows
LicenceApache-2.0
UpdatedEvidence collected 26 Sept 2026

How the rating was worked out

Weights for Coding agents: adoption 35%, experts 40%, crowd 25%. Fewer than two pillars have enough evidence. The method.

Adoption & momentumweight 35%90GitHub stars 68k → 86 · 20 releases in 90 days → 100
reach × 0.7 + momentum × 0.3
Expertsindependent reviews · weight 40%—No independent reviews found
Crowdcommunity sample · weight 25%—Only 1 community item sampled (6 needed)
Rating—Not rated: fewer than two pillars have enough evidence.

Evidence checks. 2 GitHub issues not counted as sentiment. How evidence is checked.

Trust & safety

Seven dimensions graded A–D from documented facts. Overall: A.

A

Permission model

Separate sandbox mode (read-only / workspace-write / danger-full-access) and approval policy (untrusted / on-request / never); docs show workspace-write with on-request approval as the configured posture, and native sandboxing on macOS, Linux and Windows. source

A

Data access scope

Permission profiles scope filesystem access to workspace roots with per-path read/write/deny rules and network domain allowlists. source

A

Data storage

Runs locally; config and session state kept under ~/.openinterpreter; local model providers keep model traffic on the machine. source

n/e

Data retention & training

No vendor-hosted data store; retention and training depend on the model provider the user connects. source

A

Incident history (24 months)

No published security advisories or CVEs found for the repository in the last 24 months; security reports route to the maintainer or OpenAI's disclosure programme for inherited Codex issues. source

D

Compliance

No SOC 2 / ISO attestation; small open-source project. source

A

Transparency

Apache-2.0; public GitHub releases, public docs and a documented list of deltas from upstream Codex. source

The experts

0 named independent reviews found; dated reviews from the last 12 months are scored.

  • None found.

The crowd

1 coded posts from hackernews.

1 positive0 mixed0 negative
  • positive
    Open Interpreter Classic (python version) for Swiss army knife terminal replacement one-off task type stuff.news.ycombinator.com (DeepSeek-V4-Flash Update) · 31 Jul 2026 · news.ycombinator.com

Pricing

As published, checked 26 Sept 2026.

PlanPriceNotes
Open sourceFreeApache-2.0; users bring a ChatGPT login, API keys or local models.

Recent changes

Ranked alongside

Other ranked entries in Coding agents.

Compare with the top two
OpenCodeAnomaly75
Claude CodeAnthropic73
Zed (Agent Panel)Zed Industries73
GitHub CopilotGitHub (Microsoft)72

Sources

Every figure above traces to one of these, observed up to 26 Sept 2026.

  1. Introducing Open Interpreter (search snippet: 'A new open-source coding agent for low-cost and open-weight models')
  2. PyPI open-interpreter (legacy Python package): 29,140 downloads last month