DeepSeek Harness gained over 60,000 GitHub stars within hours of its August 2026 launch, establishing itself as the first fully modular open-source agent runtime where literally every component is a swappable plugin. Built on the Cordis framework—a programming paradigm for spatiotemporal composability—dsh decomposes the entire agent stack into independently replaceable pieces: model adapters for DeepSeek, Anthropic, OpenAI, AWS Bedrock, Azure, and Google Gemini; tool registries covering bash execution, file system operations, web search, subagent delegation, and todo management; plus session stores, sandboxes, approval policies, orchestration loops, and the user interface itself. Four operating modes serve different workflows: Standard provides the full toolset, Code mode uses model-generated code to compose multi-round tool calls, Minimal strips down to a shell and editor for benchmarking, and Creator mode lets developers inspect the running runtime and test Cordis plugins in memory. The kernel handles plugin mounting, unmounting, and dependency resolution while typed events and services coordinate between components. Profiles and bundles allow the same codebase to produce entirely different products—a terminal coding agent, a browser-based workspace, a headless automation service, or an ACP/JSON-RPC endpoint—by swapping YAML configuration layers. Session history is stored as an append-only event stream for full trajectory replay, and project-level hooks on agent lifecycle events enable fine-grained behavioral customization. MCP client integration connects to external tool servers, while Agent Client Protocol enables programmatic orchestration. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Instead of babysitting three coding agents in three terminal windows, you describe the work once and let Mikka run the team: its Coordinator bot plans the job, dispatches runs to Claude Code, Codex, or Cursor on your connected machines, and only interrupts you when a decision is actually yours. Every piece of work lives in a thread with its own conversation, so dispatched runs appear as cards that report progress, ask questions, and flag blocks while you watch, reply, or stop them from the web, desktop, or phone. Unclear requests come back as choice cards before anything runs, and finished work moves through review and verification runs, back to a fix run when a check fails, then to acceptance where the pull request merges as you. Runs use leases and checkpoints, so a machine that drops mid-task hands the work to another connected computer with the same repository. You can create your own bots beside the Coordinator, each with its own chat, memory, and threads, and answers can arrive as charts, tables, stat cards, and task chips with sources cited. Daemons connect over SSH or from the desktop app, other agents drive Mikka through MCP, and Telegram or Slack messages reach the Coordinator too. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. AGPL-3.0 licensed.

Open-sourced by the Databricks AI team under Apache 2.0 and reaching over 8,500 GitHub stars within two months of launch, Omnigent introduces the meta-harness concept: a common orchestration layer that sits above existing AI coding agents and makes them interoperable parts of a governed, collaborative system. The platform wraps Claude Code, Codex, Cursor, OpenCode, Hermes, Pi, and any custom agent defined in a simple YAML configuration file into sandboxed sessions with a uniform API, then exposes each session through the terminal, a web UI, a native desktop application, mobile interfaces, and a REST API. Built-in multi-agent workflows include Polly, a coding orchestrator that delegates tasks to parallel sub-agents in separate git worktrees and routes each diff to a reviewer from a different vendor, and Deep Research, which plans sub-queries, searches the live web through MCP servers, reads full pages, and cross-checks claims across independent sources. Contextual security policies go beyond static allow/deny rules by maintaining per-session state to enforce spend caps, model routing, approval gates for destructive actions, PII blocking, and repository-scoped write restrictions across server-wide, per-agent, and per-session levels. The OS sandbox restricts filesystem and network access while intercepting egress requests to inject credentials only on approved calls. Cloud sandbox providers including Modal, Daytona, E2B, CoreWeave, Kubernetes, and Databricks launch disposable execution environments per session. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache 2.0 licensed.

Ogcode puts a full AI coding agent in a browser tab pointed at your own server, where it maps your codebase semantically before reading it, plans changes across files, runs your builds and test suites, and answers with diagrams and documents alongside the shipped code. Before reading files it maps the project with a semantic index, outlines what matters with exact line ranges, and reads only the parts that count, so unfamiliar code becomes navigable in one session. Ask it to implement a feature or chase a failing test and it plans the work, edits across files, runs your builds and test suites, and investigates failures instead of retrying blindly. You stay in control through per-session permission modes: Ask approves each step, Auto gates by risk, and Yolo lets it run, with every decision remembered when you want it to be. Past decisions are recalled instead of replaying whole transcripts, so long sessions stay focused and affordable, and the agent can research as it goes with built-in web search, reusable SKILL.md workflows, and external MCP servers. Answers come back richer than text when the work calls for it, with diagrams, rendered documents, charts, and downloadable artifacts in the same session. Connect Anthropic, OpenAI, OpenRouter, or a local Ollama endpoint and start working. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. AGPL-3.0 licensed.

Trinity runs your AI agents the way a company runs employees, with departments, roles, reporting lines, and receipts for everything they do. Each agent holds a seat on record tied to an objective and a declared metric, so whether it is working is a number with a date on it rather than an impression. Agents work on three runtimes, Claude Code, OpenAI Codex, and Gemini CLI, chosen per chat without changing the agent's state, history, or permissions. A schedule fires, the agent works inside its own container, and anything irreversible waits in a human approver's Inbox as a card naming the agent, the skill, and the input until someone approves, discusses, or rejects it. Every agent is a git repository you own: it commits its own work, its tool calls land in an append-only audit log, and every run appears in an execution ledger with its trigger, cost, and outcome, searchable across the fleet. The same agents answer in the browser Workspace, in Slack, Telegram, and WhatsApp, over a public chat link, through MCP, and from the CLI. Guardrails bind agents by what they hold rather than what they are told, with hooks they cannot switch off, cost limits, loop budgets, and a kill switch for outbound calls. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache-2.0 licensed.

Aizen is a coding agent that actually lives on your server: one small binary you can message from your phone, which fixes code, runs the tests, and reports back before you are anywhere near a keyboard. Leave it running on the box and drive it from Telegram or Discord, approving risky edits with a tap while it works behind NAT with no inbound port to expose. In the terminal it is a single chat and agent loop with a live heads-up display of model, tokens, and context, reading files in parallel, editing symbolically through the language server, and running your shell under an approval layer with a hard floor that refuses catastrophic commands even on auto-approve. It does not get to claim done early: a verify gate runs your typecheck and test suite and sends it back to fix what it broke, git-backed checkpoints let you rewind a bad turn, and seven capability-scoped sub-agents can fan out across finding, planning, building, reviewing, testing, research, and recall when a job wants a team. Point it at any OpenAI-compatible endpoint, from hosted models to your own local server, extend it with MCP servers and slash-command macros, and let the sandbox keep every child process away from your API keys with the network denied by default. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache-2.0 licensed.

Clay Studio is a self-hosted workspace where your team and its coding agents share projects, live sessions, and project knowledge in one browser interface instead of scattering them across terminals. You organize several repositories in one sidebar and run sessions in parallel while teammates watch live activity, enter the same session, hand work off, and answer permission requests from wherever they are. Ten agent runtimes plug into the same workspace, Claude Code, Codex, Grok Build, Kimi Code, GitHub Copilot CLI, Qwen Code, Junie CLI, Antigravity CLI, OpenCode, and Kiro CLI, each keeping its own authentication and models. Persistent Mates give AI collaborators an identity, instructions, memory, direct messages, and roles in structured debates that survive beyond a single chat. Worktrees, paired sessions, spawned workers, scheduled tasks, and Ralph Loops let background jobs proceed without losing sight of what is running. You browse and edit files, inspect diffs and history, open terminals, install skills, connect MCP servers, and manage project knowledge without leaving the workspace. The installable PWA works on mobile and pushes notifications for approvals, failures, and finished work. Sessions and knowledge stay on your disk as portable JSONL and Markdown, so switching providers never means migrating your workspace. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. AGPL-3.0-only licensed.