llms.txt
Compare with other sites' llms.txt files →# Omnigent
> Omnigent is an open-source meta-harness for AI agents. It provides a common layer over Claude Code, Codex, Pi, and custom agents, with shared sessions, contextual policies, local and cloud sandboxing, model and credential configuration, real-time collaboration, and multiple interfaces.
Use this file to find the most relevant Omnigent documentation for coding agents, custom agents, deployment, collaboration, policies, sandboxing, and end-user interfaces.
For a single-file expanded context, use [llms-full.txt](https://omnigent.ai/llms-full.txt).
## Quickstart
- [Tutorial: Coding Agent](https://omnigent.ai/quickstart/coding-agent): Tutorial: see what Omnigent adds on top of Claude Code or Codex, including a web UI, mobile access, session sharing, and forking a session into a different agent.
- [Tutorial: Collaborate from Anywhere](https://omnigent.ai/quickstart/collaborate): Tutorial: deploy a server to access agents from your phone, share live sessions with teammates, and run agents on cloud sandbox hosts.
- [Install](https://omnigent.ai/quickstart/install): Install the Omnigent CLI with the installer script, or download the desktop app.
- [Tutorial: Contextual Policies](https://omnigent.ai/quickstart/policies): Tutorial: attach context-aware policies to a live session to warn on spend, pause for approval, or block risky actions, with no code required.
- [Tutorial: Polly (Multi-AI Coding)](https://omnigent.ai/quickstart/polly): Tutorial: see Polly in practice, a multi-agent coding orchestrator that delegates sub-tasks across AI agents with built-in cross-vendor review.
## Agent Use Cases
- [Built-in Multi-AI Agents](https://omnigent.ai/docs/use/builtin-agents): Omnigent ships with multi-agent orchestrators that combine different AI agents into a single workflow, ready to run out of the box.
- [Debby](https://omnigent.ai/docs/use/builtin-agents/debby): Debby asks two models the same question and runs a multi-round debate where each critiques and refines the other's answer.
- [Polly](https://omnigent.ai/docs/use/builtin-agents/polly): Polly is a supervisor that delegates sub-tasks to sub-agents on separate harnesses and worktrees, with cross-agent review and human-decided merges.
- [Coding Agents](https://omnigent.ai/docs/use/coding-agents): Run Claude Code or Codex through Omnigent with one command and get a web UI, persistent sessions, collaboration, mobile, and policies.
- [Custom Agents](https://omnigent.ai/docs/use/custom-agents): Define a custom agent in a short YAML file, with no code or frameworks, and get the full Omnigent platform around it.
## Harnesses
- [Harnesses](https://omnigent.ai/docs/build/harnesses): Understand the runtime layer that executes your agent loop and how to launch or select a harness.
- [Community Harnesses](https://omnigent.ai/docs/build/harnesses/community): Package and register a new Omnigent harness as a community plugin.
- [Configuration](https://omnigent.ai/docs/build/harnesses/configuration): Configure built-in and custom ACP harnesses, authentication, runtime selection, and launch overrides.
- [Supported Harnesses](https://omnigent.ai/docs/build/harnesses/supported): See every harness available in Omnigent, including its ids, aliases, execution modes, and support status.
## Build Custom Agents
- [Models & Credentials](https://omnigent.ai/docs/build/models): Declare your model in agent YAML and let Omnigent manage the credentials that authenticate to each provider.
- [Prompts & Skills](https://omnigent.ai/docs/build/prompts): Shape agent behavior with system prompts, and extend it with skills: reusable instruction sets the agent loads on demand.
- [Smart Routing](https://omnigent.ai/docs/build/routing): Let a router pick the best harness and model for each session automatically, using the built-in judge or an external routing API.
- [Scheduled Tasks](https://omnigent.ai/docs/build/scheduled-tasks): Run an agent on a recurring schedule. A scheduled task is a saved prompt that fires an agent session on an RRULE cadence until you delete it.
- [MCP & Tools](https://omnigent.ai/docs/build/tools): Give your agent capabilities beyond conversation. Declare tools and MCP servers in your agent YAML, each with a name you choose.
## Interfaces
- [Desktop App](https://omnigent.ai/docs/interact/desktop): The Omnigent desktop app brings the full web UI to a native window, with notifications, a dock badge, and multiple windows.
- [Mobile](https://omnigent.ai/docs/interact/mobile): Use Omnigent on your phone: native apps for iOS and Android, or the mobile web UI in any phone browser. Start a session on your laptop, continue it on your phone.
- [Slack](https://omnigent.ai/docs/interact/slack): Talk to your Omnigent agents from Slack: set up the bot, map threads to sessions, stream replies, and approve tool calls without leaving your workspace.
- [Terminal](https://omnigent.ai/docs/interact/terminal): Omnigent's keyboard-first default interface, with streaming output and real-time tool execution for people who live in the terminal.
- [Web UI](https://omnigent.ai/docs/interact/web-ui): The Omnigent web UI starts automatically with every session; open the printed URL in any browser to work with your agent.
## Collaboration And Deployment
- [Pair Programming](https://omnigent.ai/docs/collaborate): Omnigent is multi-user: share a live session by link, let a teammate co-drive, or fork a conversation across terminal, web, and mobile.
- [Auth & SSO](https://omnigent.ai/docs/collaborate/auth): Authenticate users in Omnigent three ways. Pick the approach that fits your setup, from simple tokens to full SSO.
- [Cloud Sandbox Host](https://omnigent.ai/docs/deploy/cloud-sandbox-host): Move the Omnigent runner from your laptop to a remote container so your agent keeps working in an isolated cloud environment.
- [Community Sandbox Providers](https://omnigent.ai/docs/deploy/community-sandbox-providers): Implement, package, and test a third-party sandbox provider for Omnigent using the launcher API and Python entry-point registry.
- [Credential Store](https://omnigent.ai/docs/deploy/credential-store): Encrypt integration credentials at rest with a pluggable cipher backend. Choose AWS KMS or HashiCorp Vault Transit per server.
- [Database](https://omnigent.ai/docs/deploy/database): The Omnigent server persists session history, accounts, and artifacts. Choose between Postgres, CockroachDB, and SQLite, with the same schema and migrations.
- [Connect Databricks](https://omnigent.ai/docs/deploy/databricks-connect): Wire a Databricks OAuth app so signed-in users can connect their workspace and their managed sandboxes reach the Databricks AI Gateway as them.
- [Connect GitHub](https://omnigent.ai/docs/deploy/github-connect): Wire a GitHub App so signed-in users can connect their GitHub account and their managed sandboxes authenticate git and gh as them.
- [Shared Server](https://omnigent.ai/docs/deploy/overview): Deploy the Omnigent server so your agents are reachable from your phone, shareable with teammates, and running while your laptop sleeps.
- [Sandbox Profiles](https://omnigent.ai/docs/deploy/sandbox-profiles): Let a sandbox provider offer per-user, provider-authorized sandbox configurations that users pick when starting or cloning a managed session.
- [Usage Telemetry](https://omnigent.ai/docs/deploy/telemetry): What anonymous usage telemetry the Omnigent server collects, and every way to turn it off.
## Policies
- [Builtin Policies](https://omnigent.ai/docs/policies/builtin): Built-in Omnigent policies for common safety and cost-control guardrails. Apply them by name or reference them in YAML.
- [Custom Policies](https://omnigent.ai/docs/policies/custom): Extend Omnigent with custom policy logic — use cel_policy for most cases, or write a Python policy when CEL's non-Turing-complete expressions aren't enough.
- [Contextual Policies](https://omnigent.ai/docs/policies/overview): Policies control what your Omnigent can do, intercepting every action to allow, ask for approval, or deny it in real time.
## Omnibox
- [Omnibox](https://omnigent.ai/docs/omnibox): Omnibox is Omnigent's secure OS-level sandbox that wraps any agent to run it unattended without exposing your credentials, files, or network.
## Programmatic Usage
- [Programmatic Usage](https://omnigent.ai/docs/programmatic): Run agents from scripts and CI, drive sessions with the Python SDK, and manage sessions through the Omnigent REST API.
- [Configure Omnigent in GitHub Actions](https://omnigent.ai/docs/programmatic/github-actions): Run Omnigent with Databricks AI Gateway in GitHub Actions for pull request review.
## Reference
- [Configuration Reference](https://omnigent.ai/docs/reference/configuration): Reference for Omnigent user, project, and server configuration.
- [Cloud Sandbox Configuration](https://omnigent.ai/docs/reference/configuration/cloud-sandbox): Reference for configuring managed cloud sandbox providers, lifecycle settings, credentials, and host configuration.
- [Kubernetes Sandbox Configuration](https://omnigent.ai/docs/reference/configuration/kubernetes): Configure storage and Secret mounts for Kubernetes managed sandbox runner Pods.
- [OS Sandbox Configuration](https://omnigent.ai/docs/reference/configuration/os-sandbox): Omnibox is Omnigent's secure OS sandbox: restrict filesystem and network access and hide credentials for unattended, locked-down agents.
- [Smart Routing Configuration](https://omnigent.ai/docs/reference/configuration/routing): Reference for enabling built-in or external smart routing and configuring the external routing service.
- [External Routing API](https://omnigent.ai/docs/reference/routing-api): Request and response contract for services that provide Smart Routing decisions to an Omnigent server.
## Answers
- [Answers](https://omnigent.ai/answers): Browse question-led explanations and practical examples.
- [What is a meta-harness?](https://omnigent.ai/answers/meta-harness): A meta-harness coordinates coding-agent runtimes. Compare manual coordination and native delegation, then see where Omnigent’s shared session tools fit.
- [What is a contextual policy?](https://omnigent.ai/answers/contextual-policy): A contextual policy uses action and session state to decide whether agent work proceeds. Compare permissions and hooks, then examine Omnigent’s policy layer.
- [How do I orchestrate Codex and Claude Code sessions together?](https://omnigent.ai/answers/orchestrate-codex-and-claude-code): Orchestrate Codex and Claude Code with terminals, native delegation, or a meta-harness. Compare the options and launch both through Omnigent’s app or web UI.
- [How do I hand off a coding agent session to a teammate?](https://omnigent.ai/answers/hand-off-a-session-to-a-teammate): Hand off a coding-agent task with a clear state brief and code changes. Compare shared sessions, conversation forks, and repository-based handoffs in Omnigent.
- [How do I schedule recurring tasks for an AI coding agent?](https://omnigent.ai/answers/schedule-recurring-agent-tasks): Schedule coding-agent work with cron, CI, or a session scheduler. Compare execution targets, permissions, run checks, and Omnigent’s recurring-task controls.
- [How do I start a coding agent task on my laptop and continue it on my phone?](https://omnigent.ai/answers/continue-a-session-on-your-phone): Continue a coding-agent task from your phone using remote terminal or session access. Compare options and configure Omnigent without moving laptop execution.
- [How do I set spending limits for coding agents?](https://omnigent.ai/answers/set-spending-limits-for-coding-agents): Set coding-agent budgets by choosing what to count and where to stop work. Compare alerts, billing controls, and Omnigent’s recorded-cost policy enforcement.
## Optional
- [Homepage](https://omnigent.ai/): Product overview and architecture.
- [FAQ](https://omnigent.ai/faq): Answers to common questions about Omnigent, models, custom agents, safety, Databricks, and production readiness.
- [Feature notes](https://omnigent.ai/feature-notes): Short posts spotlighting individual Omnigent features.
### Blog
- [Review, label, and merge your agents' pull requests without leaving the session](https://omnigent.ai/blog/review-merge-agent-prs-in-session): The Pull Requests tab now acts on GitHub pull requests: edit the title and description, add labels, request reviewers, post comments, open CI runs, and merge, all beside the agent's conversation.
- [Install a custom agent once from GitHub and use it on every host](https://omnigent.ai/blog/install-custom-agents-from-github): Install a custom agent with omnigent agent add from a local directory, a bundle, or a GitHub repo. It is stored under your account and shows up in the new-session picker on every host you use.
- [Read your agent's pull or merge request in the session](https://omnigent.ai/blog/github-pr-review-in-session): Read GitHub and Azure DevOps pull requests or GitLab merge requests alongside your agent's conversation. Inspect descriptions, comments, CI results, and file diffs in Omnigent's read-only Pull Requests tab.
- [Moving agent orchestration from code to configuration](https://omnigent.ai/blog/declarative-agent-configuration): Define the orchestration setup you want in YAML, and let Omnigent handle the sessions, messaging, isolation, and policy enforcement that make it run.
- [Stop Surprise LLM Bills With Production-Ready Cost Controls](https://omnigent.ai/blog/stop-surprise-llm-bills): How Omnigent keeps LLM spend predictable: smart routing, progressive budgets, custom pricing for self-hosted models, and per-model usage tracking.
- [Give your agents built-in structured web research and extraction](https://omnigent.ai/blog/nimble-web-research-extract-builtins): Enable nimble_research and nimble_extract by name in your agent spec: your agents can run cited web research and pull structured data off live pages, no glue code required.
- [Allow, ask, deny: contextual policies in Omnigent](https://omnigent.ai/blog/omnigent-policies): An introduction to Omnigent policies: contextual ALLOW/ASK/DENY guardrails for AI agents, with built-in, custom Python, and CEL examples.
- [Meta-Harness: the Missing Layer Above Claude Code, Codex, and Pi](https://omnigent.ai/blog/meta-harness-missing-layer): Use Claude Code, Codex, Pi, and other coding agents together through one shared orchestration, collaboration, and policy layer.
- [Organize your sessions into projects, with optional defaults that pre-fill every new chat](https://omnigent.ai/blog/first-class-projects-with-defaults): Projects keep related sessions together instead of scattered across a flat list. Each one can optionally carry default session settings (host, working directory, agent, and a fresh git worktree) so new chats open…
- [Plug any ACP agent into Omnigent with your tools and governance built in](https://omnigent.ai/blog/any-acp-agent-with-tools-and-governance): Point Omnigent at any agent that speaks the Agent Client Protocol and run it with your session tools, sub-agents, skills, and policies already wired in.
- [Line up your next moves while an agent works — steer or reorder them mid-run](https://omnigent.ai/blog/queue-and-steer-messages-while-agents-work): Type follow-ups while a turn is running: they queue in a strip above the composer and send in order when the agent goes idle, or you can steer, edit, reorder, and delete them.
### Releases
Release notes for all 21 versions are in [llms-full.txt](https://omnigent.ai/llms-full.txt); the newest are summarized here.
- [Omnigent v0.21.0](https://omnigent.ai/releases/0.21.0): Released 2026-11-05. What's new in Omnigent v0.21.0 — release highlights, improvements, and fixes.
- [Omnigent v0.20.0](https://omnigent.ai/releases/0.20.0): Released 2026-10-28. What's new in Omnigent v0.20.0 — release highlights, improvements, and fixes.
- [Omnigent v0.19.1](https://omnigent.ai/releases/0.19.1): Released 2026-10-21. What's new in Omnigent v0.19.1 — release highlights, improvements, and fixes.
- [Omnigent v0.19.0](https://omnigent.ai/releases/0.19.0): Released 2026-10-20. What's new in Omnigent v0.19.0 — release highlights, improvements, and fixes.
- [Omnigent v0.18.0](https://omnigent.ai/releases/0.18.0): Released 2026-10-13. What's new in Omnigent v0.18.0 — release highlights, improvements, and fixes.
- [Omnigent v0.17.0](https://omnigent.ai/releases/0.17.0): Released 2026-10-06.
- [Omnigent v0.16.0](https://omnigent.ai/releases/0.16.0): Released 2026-09-29.
- [Omnigent v0.15.0](https://omnigent.ai/releases/0.15.0): Released 2026-09-24.
- [Omnigent v0.14.0](https://omnigent.ai/releases/0.14.0): Released 2026-09-15.
- [Omnigent v0.13.0](https://omnigent.ai/releases/0.13.0): Released 2026-09-09.
- [Omnigent v0.12.0](https://omnigent.ai/releases/0.12.0): Released 2026-09-01.
- [Omnigent v0.10.0](https://omnigent.ai/releases/0.10.0): Released 2026-08-19.
- [Omnigent v0.9.0](https://omnigent.ai/releases/0.9.0): Released 2026-08-11.
- [Omnigent v0.8.0](https://omnigent.ai/releases/0.8.0): Released 2026-08-03.
- [Omnigent v0.7.0](https://omnigent.ai/releases/0.7.0): Released 2026-07-27.
- [Omnigent v0.6.0](https://omnigent.ai/releases/0.6.0): Released 2026-07-21.
- [Omnigent v0.5.1](https://omnigent.ai/releases/0.5.1): Released 2026-07-10.
- [Omnigent v0.5.0](https://omnigent.ai/releases/0.5.0): Released 2026-07-10.
- [Omnigent v0.4.0](https://omnigent.ai/releases/0.4.0): Released 2026-07-03.
- [Omnigent v0.3.0](https://omnigent.ai/releases/0.3.0): Released 2026-06-27.
- [Omnigent v0.2.0](https://omnigent.ai/releases/0.2.0): Released 2026-06-19.
### Other
- [GitHub repository](https://github.com/omnigent-ai/omnigent-site): Website source code.
- [Sitemap](https://omnigent.ai/sitemap.xml): Full public route index.