Skip to content

Omnigent Review

Omnigent is a free, open-source (Apache 2.0) meta-harness from the Databricks AI team that runs Claude Code, Codex, Pi and your own agents under one layer of policies, sandboxing and shared live sessions.

Free · API

Stork's verdict on Omnigent

Omnigent is a free, fast-moving way to put one set of policies and spend caps over several coding agents, best for teams willing to run pre-1.0 software.

Omnigent reviewed by Stork AI · stork.ai/en/omnigent

stork
I am one still image.
One typed sentence. This clip cost 5 cents, the minimum per video.
Make a photo talk for 20 cents a minute.
Upload one photo and type the words. No subscription. One free video a day.
Make one freeDevelopers: there is an API

What Omnigent does

Omnigent (omnigent.ai) calls itself a meta-harness: a common layer that sits above coding agents such as Claude Code, Codex and Pi, plus agents you write yourself in YAML. A runner wraps any agent in a uniform session that can run in a local OS sandbox or a remote cloud sandbox, and a server adds policies and shared history and exposes every session over the terminal, a web UI, desktop apps, iOS and Android apps, Slack and a REST API. The site says it is "Built by the Databricks AI team, Neon and the Omnigent Contributors" and licensed Apache 2.0; Databricks announced it on June 13, 2026 as open source in alpha. It is still pre-1.0: the current release is v0.21.0 (November 5, 2026), with releases roughly weekly since June.

Setup is a one-line install (curl -fsSL https://omnigent.ai/install.sh | sh) that needs Python 3.12+, Node.js 22 and tmux, then omni setup to add model credentials: API keys (Anthropic, OpenAI, OpenRouter, Groq, DeepSeek, xAI, Mistral, Together AI, Fireworks), an existing Claude Pro/Max or ChatGPT Plus/Pro login, an OpenAI- or Anthropic-compatible gateway (LiteLLM, Portkey, Ollama, vLLM and others) or a Databricks workspace. omni claude or omni codex then runs your usual agent with Omnigent's web UI, persistent sessions and policies on top. For a team you deploy the server (Docker Compose with Postgres, or a hosted platform) so sessions survive a sleeping laptop and teammates can join them.

Omnigent pricing

PlanMonthly priceIncluded
Omnigent (open source, self-hosted)$0Everything: CLI, server, web UI, desktop and mobile apps, built-in agents, policies, Omnibox sandbox, REST API. Apache 2.0 license
Omnigent on Databricks (managed, Beta)Not publishedDatabricks-operated server tied to your workspace identity, models through Databricks Foundation Model APIs and Unity Gateway, Databricks Sandboxes

Omnigent has no pricing page and no paid tier of its own. What you pay for is what it orchestrates: model usage through your own API keys or gateway (or an existing Claude Pro/Max or ChatGPT Plus/Pro subscription, which Omnigent can use through the official claude and codex CLIs), any cloud sandbox provider you add (Modal, Daytona, E2B, Kubernetes and others) and wherever you host the server. Its cost-control policies exist to cap that model spend, per session, per sub-agent or per user per day, week, month, quarter or year.

The managed Omnigent on Databricks is in Beta, needs a workspace admin to enable the Omnigent preview, and requires a workspace in a region that supports Unity Gateway. Databricks' Omnigent docs do not state a price for it. On the managed version, sandbox hosts route models through Unity Gateway, so you cannot bring your own model API keys, and custom Python policies are not supported (built-in and CEL policies are).

The server collects anonymous usage telemetry by default since v0.6.0; set OMNIGENT_ANALYTICS=0 or DO_NOT_TRACK=1 to turn it off.

Prices checked on the vendor's pricing page on November 10, 2026. Omnigent publishes no pricing page; see the Omnigent install docs, the Apache 2.0 repository and Omnigent on Databricks.

Key features

What Omnigent's own docs and release notes list:

  • Harnesses: Claude Code, Codex and OpenAI Agents SDK fully supported; Copilot, Cursor, OpenCode and Pi maintained; Antigravity, Devin, Hermes, Kimi, Kiro, Qwen Code and others community-supported; Goose via ACP
  • Polly, a built-in orchestrator that splits a task across Claude Code, Codex, Cursor, Pi and other sub-agents in separate git worktrees with cross-vendor review
  • Debby, a built-in agent that sends every question to both Claude and GPT and lets them debate
  • Contextual policies that return ALLOW, ASK or DENY with session state: cost budgets, tool-call caps, loop detection, PII blocking, GitHub force-push blocking, model routing
  • Omnibox OS sandbox on Linux (bubblewrap, seccomp) and macOS (Seatbelt): filesystem and network allow-lists, with credentials injected by a proxy so the agent never reads them
  • Live sessions shared by URL with full history; teammates can comment and steer from the web UI, desktop apps (macOS, Linux, Windows), iOS and Android apps or Slack
  • Import existing Claude Code and Codex chats as Omnigent sessions (up to 100 at a time)
  • Scheduled automations, a REST API, GitHub Actions usage and a Pull Requests panel to review and merge an agent's PRs in the session

Who Omnigent is best for

Omnigent suits engineers and platform teams who already run more than one coding agent and want one place to put them: the same policies and spend caps over Claude Code and Codex, a sandbox they can leave running unattended, and sessions a teammate can pick up from a browser or phone. It is also a fit for Databricks shops that want the managed Beta tied to workspace identity and AI Gateway. Expect pre-1.0 software: frequent releases, community-supported harnesses that the team fixes less urgently, and an OS sandbox that only covers Linux and macOS.

If you want to build multi-agent applications in code rather than wrap existing coding agents, look at CrewAI or LangGraph. If you want a single open-source coding agent rather than a layer above several, compare OpenHands or Goose, which Omnigent can itself run as a harness.

What users say

Independent hands-on reviews were scarce at the time of writing, and no G2 or Capterra listing turned up. Coverage so far is launch reporting. heise online (June 17, 2026) describes Omnigent as early-stage, advises trying it first in less critical environments, and notes that Databricks has not documented reference deployments at the scale of hundreds of agents. OpenSourceForU (June 19, 2026) also calls it early in its lifecycle and relays Databricks' feature claims without independent testing. The GitHub repository is the best place to judge activity and open issues before adopting it.

Pros

  • Free and open source under Apache 2.0, with no paid tier to unlock features
  • One policy and spend-cap layer over Claude Code, Codex and many other agents
  • Kernel-enforced sandbox that hides real credentials from the agent
  • Same live session from terminal, browser, desktop, phone or Slack
  • Can use existing Claude Pro/Max or ChatGPT Plus/Pro logins

Cons

  • Pre-1.0 software announced as alpha in June 2026, releasing roughly weekly
  • Omnibox OS sandbox covers Linux and macOS only
  • Many harnesses are community-supported, with slower fixes
  • Shared team use means hosting and maintaining your own server

Frequently asked questions

Is Omnigent free?

Yes. Omnigent is open source under the Apache 2.0 license and has no paid tier of its own. You pay for the models it calls (your own API keys, a gateway, or a Claude Pro/Max or ChatGPT Plus/Pro login), any cloud sandboxes you add, and hosting for a shared server. The managed Omnigent on Databricks (Beta) has no published price.

Who makes Omnigent?

The Omnigent site says it is built by the Databricks AI team, Neon and the Omnigent Contributors. Databricks announced it on June 13, 2026 and also offers a managed version, Omnigent on Databricks, in Beta.

What is a meta-harness?

A harness is the runtime that executes an agent loop, such as Claude Code or Codex. Omnigent wraps those harnesses in a common session, so you can switch or combine them, apply the same policies and sandboxing to all of them, and share the live session with teammates.

Which coding agents does Omnigent support?

Its docs list Claude Code, Codex and the OpenAI Agents SDK as fully supported; Copilot, Cursor, OpenCode and Pi as maintained; and Antigravity, Devin, Hermes, Kimi, Kiro, Qwen Code, Rovo Dev and ZCode as community-supported. Goose, Grok Build and Jcode run through its ACP integration.

Can Omnigent use my Claude Pro or ChatGPT subscription?

Yes. Omnigent can use an existing Claude Pro/Max or ChatGPT Plus/Pro plan through the official claude and codex CLIs, with no API key needed. It also accepts API keys from providers such as Anthropic, OpenAI and OpenRouter, or any OpenAI- or Anthropic-compatible gateway.

How do I set spending limits for coding agents in Omnigent?

Add a cost policy. Built-in policies cap LLM spend per session (asking at soft thresholds, blocking expensive models at the hard cap), per sub-agent, or per user per day, week, month, quarter or year. You can add one by asking the agent in chat, in the agent YAML, or server-wide for every session.

Does Omnigent work on Windows?

Desktop app installers exist for macOS, Linux and Windows. The Omnibox OS sandbox only runs on Linux and macOS, release notes for v0.3.0 described native Windows support as core features only, and launching SSH sessions from the desktop app is not available on Windows yet.

Compare Alternatives

See every Omnigent alternative, compared

Omnigent

Free · API

From the Stork blog

Enjoying this? Get one like it in your inbox each morning.

one email a day · unsubscribe in two clicks · no third-party tracking