Omnigent: How to Run Many Agent CLIs Together
See how Omnigent controls Claude Code, Codex and other agent CLIs, pin release 0.7.0, and sandbox each agent for VPS use while e still dey alpha.
Wetin Omnigent be
Omnigent na open source meta-harness: one orchestration layer wey dey drive the agent command line tools (CLIs) wey you don already install. E no replace Claude Code, Codex, Cursor, OpenCode, Hermes or Pi. E start dem, give each one job, and supervise the result inside one session with one policy set. Databricks publish the repository for June 2026 under Apache 2.0 licence, and PyPI still classify the package as alpha.
The practical claim no wide, and e worth explain am plainly. You describe one agent once for YAML, then name the harness wey go run am. Change that one line, and the same agent go run for another vendor CLI. Nothing else for your setup go change, because Omnigent dey control the loop above the agents, instead of the loop inside dem.
Meta-harness na wetin, and how e different from framework?
Harness na program wey wrap model inside loop. E dey read your prompt, call tools, edit files, and report result back. Claude Code na harness. Codex na harness. You install am, log in, and e dey work by itself.
Framework na library wey you write code against. You import am, define the steps for Python, and your program become the agent. If you change vendor for there, you need edit your code, because vendor client dey wired through your program.
Meta-harness dey one level above both of dem. E na supervisor wey dey run harnesses as child processes. Omnigent dey start the vendor CLI, give am work, and read wetin come back. You keep the CLI wey you don already install, and you keep the subscription or API (application programming interface) key wey already dey pay for am. Na this be the whole difference, and na this decide who the tool dey for: people wey get several agent CLIs wey already dey work, but don tire to drive dem one terminal at a time.
Orchestration layer wey dey solve which problem?
- Changing vendor na one line. Agent definition dey keep
harnessandmodelas data, so to move one role from one vendor go another na edit for the YAML file, no be rewrite. - Review fit cross vendors. Model from different company fit read diff wey another model write. Two models from the same family dey usually get the same blind spots, so second opinion from the same vendor no get the same value.
- Policy get one home. Agent file dey declare budget thresholds and approval prompts, and dem apply to every sub-agent underneath am.
- Session fit outlive any one tool. One transcript dey cover work wey several CLIs do, so you fit read wetin happen without joining four scrollbacks together.
The cost na the layer itself. Every bug for Omnigent don become bug wey dey sit between you and agent wey bin dey work by itself before. For alpha, na real cost be this, no be theoretical one.
Where multi-agent harness fit sit beside single agent tools
If you never run one agent for server before, start from there first. Our guide to running a coding agent on a VPS cover the single-agent setup from beginning to end, and na this setup Omnigent assume say you don get already. The wider field of self-hosted AI agents na where you choose the agents themselves, while learning how agents actually work better be your first stop if these terms still new to you.
Omnigent dey operate for another level from connector layer too. Work like giving agents access to your own data sources dey about the data sources wey agent fit reach. Omnigent dey about which agent go run, the order wey dem go run, and the limits wey dem go get. You fit need both at the same time, and dem no overlap.
Wetín you need before you install
- Python 3.12 or newer. The published package declares
requires-python >= 3.12. tmux, because the terminal harnesses dey run inside am.- At least one vendor CLI, wey you don install and log in already.
- Node.js 22 or newer with
npm, because the Claude, Codex and Pi harnesses need dem. The installer dey check for Node 22.10 or later and go warn if e no dey. Omnigent itself no need Node: the wheel for PyPI dey ship with the built web assets, and na only build from a git checkout dey make Node a hard requirement.
Install pinned release, no be main
curl -fsSL https://raw.githubusercontent.com/omnigent-ai/omnigent/main/scripts/install_oss.sh | sh -s -- --version 0.7.0The sh -s -- part no be decoration. If e no dey there, sh go read --version as e own option, and installer no go run at all. When sh na Bash, e go print Bash own version and exit successfully. So nothing go install, and nothing go warn you. With -s --, sh go read the script from the pipe and pass --version 0.7.0 through to the installer. For repository wey dey release breaking changes every few weeks, na this one dey separate reproducible box from surprise.
Release 0.7.0 still dey published for PyPI as of 29 October 2026, but the newest release now na 0.22.0, from 13 November 2026. The earlier 0.19.1 release na security-focused patch with no new features. E update dependencies after the eMudhra intermediate certificate authority compromise, and the project recommend say make you upgrade. Wetin matter na the pin, no be the number: pin the version wey you don test, and read the breaking changes for release notes before you move am. Pin no be reason to skip security release, though, so move am when one land.
The installer dey use uv, Astral's Python package manager, and e go offer to install uv first if e no dey. If uv don dey already, skip the script:
uv tool install --force --python 3.12 "omnigent==0.7.0"Extras dey follow the same pattern, and the flag dey repeat: --extra e2b --extra kubernetes for the script, or "omnigent[e2b,kubernetes]" with uv. The script also add the databricks extra unless you pass --no-databricks, while the uv line above install the package without am. Note say the git tag na v0.7.0 while the package version for PyPI na 0.7.0.
uv dey put the binary for the directory wey uv tool dir --bin report, usually ~/.local/bin, and the installer go offer to append am to your shell profile. If command no dey found immediately after clean install, na why. Check wetin you end up with:
omni upgrade --checkThat one compare the installed version with the latest published one and tell you whether upgrade dey available, without doing am. omni and omnigent na the same program under two names.
Point am to a model provider
omni setupThe wizard go check credentials wey already dey your environment, then ask for the ones wey dey missing. E handles API keys, vendor subscriptions, gateways like OpenRouter or Ollama, and Databricks workspaces. If you already dey run local model server with Ollama for the same machine, point one gateway to am, and the traffic no go ever leave the box.
A minimal multi-agent run
The example agents dey inside repository, so clone the same tag wey you install instead of main.
git clone --depth 1 --branch v0.7.0 https://github.com/omnigent-ai/omnigent.git
cd omnigent
omnigent run examples/polly/Polly na the multi-agent coding orchestrator wey dey ship with the repo. E config define sub-agents wey dem name claude_code, codex, opencode, cursor, hermes and pi. E get one rule wey make the whole exercise useful: na always different vendor go do review from the implementer. Polly no dey write code by itself. E dey plan, divide the goal into work items, delegate each one, and route each diff go reviewer from another vendor.
Before Polly delegate anything, e dey run preflight check to see which sub-agent CLIs actually dey available for the machine. If na only one vendor CLI dey installed, nobody dey available to review the diff. So install at least two before you assess the output. Debby, the other example wey ship with the repo, na debate agent wey get two heads: one Claude and one GPT:
omni debbyNa short way to confirm say two providers dey configured, because e need both of dem before e fit talk anything.
Declare sub-agents as tools
The agent file na YAML. executor dey name the harness, the model, and the authentication. tools dey hold MCP (model context protocol) servers, Python functions, and sub-agents. A sub-agent na tool wey get type: agent and e get its own executor. Na this mechanism dey power everything wey dey above.
name: orchestrator
prompt: |
You coordinate coding and review tasks.
executor:
harness: claude-sdk
model: databricks-claude-sonnet-4-6
tools:
coder:
type: agent
prompt: Write and test code.
executor:
harness: claude-sdk
model: databricks-claude-opus-4-7
reviewer:
type: agent
prompt: Review proposed changes.
executor:
harness: claude-sdk
model: databricks-claude-sonnet-4-6omnigent run path/to/my_agent.yamlThose model ids come from the project own docs/AGENT_YAML_SPEC.md example, and dem na names wey Databricks host. Replace harness and model with whatever omni setup configure for your box. Other harness values for the spec include antigravity, copilot, kimi and qwen. The current spec for main also document acp:<slug> for any Agent Client Protocol server, but the v0.7.0 spec no list am. So confirm say e work for the version wey you pin before you depend on am. The spec also support pass_history: true for a sub-agent. This one pass the parent conversation give am. E dey cost tokens for every delegation, so leave am off for sub-agents wey only need the task wey dey in front of dem. If coder prompt tell am to make the smallest change wey go work, e go give the reviewer a diff wey short enough to actually read. For this work, that one matter pass the model wey you choose for either role.
Why long-running orchestration belong for VPS
A multi-agent run no be two-minute command. You need plan, delegate, wait for parallel git worktrees, review, and revise. If you close laptop lid, everything go stop. VPS (virtual private server) dey online and e keep network connection, so session go continue while you no dey watch am.
omnigent server --background
omnigent server statusThe server dey host web user interface for port 6767. omnigent server status dey report whether one dey run, while omnigent stop dey stop am. v0.7.0 remove omni server start and replace am with omni server --background; v0.9.0 bring am back as deprecated alias.
No publish 6767 for public address. Two safe options dey. Keep the port closed for firewall and forward am through SSH with ssh -N -L 6767:localhost:6767 you@your-server, then open web interface at http://localhost:6767 for your own machine. Or terminate TLS (transport layer security) in front of am and turn authentication on:
OMNIGENT_AUTH_ENABLED=1 omnigent server --backgroundThe firewall part na ordinary work, and these ufw firewall basics for VPS cover am. If the box already dey run containers behind Traefik in front of several Docker Compose apps, Omnigent na just one more service for the same pattern.
For container deploy, the repository's deploy/docker/ directory get a Compose setup: ./bootstrap.sh dey mint secrets into .env, then docker compose up -d start Omnigent and Postgres, with the server for port 8000 instead of 6767. OMNIGENT_PORT change that host port. Compose compute DATABASE_URL for you. For other deploy targets, the same variable dey select Postgres or SQLite. OMNIGENT_AUTH_ENABLED default to 1 inside the containers, and that na the correct default for anything wey people fit reach from outside.
For sizing, the deploy notes put the server idle memory use around 275 MB RSS, while the Fly.io config set the machine to 1 GB. The idle measurement na for supervisor alone. The 1 GB na the memory wey the machine get allocated. Sub-agents fit share the parent's working directory and environment, including its sandbox through os_env: inherit. Separate git worktrees na orchestration choice, so size the box based on the harnesses and agents wey you run at the same time. Once server dey up, omnigent login https://your-host followed by omnigent host https://your-host register your laptop with am, while omnigent attach <session_id> pick up a running session again from another device.
Sandbox every sub-agent before you walk away
Omnigent dey ship operating system level sandbox wey dem call Omnibox. For Linux, e dey use bubblewrap namespaces plus seccomp, so na kernel dey enforce the boundary instead of the agent prompt. Agent wey prompt injection don affect no fit talk im way comot from kernel rule. Install the dependency first:
sudo apt install bubblewrapConfiguration dey under os_env for the agent file:
os_env:
type: caller_process
cwd: .
sandbox:
type: linux_bwrap
write_paths: [.]
write_files: []
read_paths: []
allow_network: true
cwd_allow_hidden: [.venv]
env_passthrough: []
egress_rules: []
credential_proxy: []Working directory dey read only until you list am for write_paths, so agent wey go wrong no fit write outside workspace. Top-level dotfiles for working directory and each read_paths root remain hidden unless you name dem for cwd_allow_hidden. This mean broad read grant no go quietly expose top-level .ssh or .aws directories. Nested dotfiles remain visible by default; set cwd_hidden_scan_recursive: true to mask dem throughout the tree. Set egress_rules and all HTTP and HTTPS traffic go through default deny proxy, with each rule written as "METHODS host/path-glob". credential_proxy go one step further: agent only ever hold placeholder, and proxy go swap am with real secret as request dey comot, so leaked transcript no leak anything wey person fit use. For multi-harness setup, each sub-agent carry its own sandbox block inside its own config file under agents/. This mean reviewer fit get network denied while implementer keep access.
Documentation state the limit, and e important. OS sandbox dey apply to sys_os_* tool calls and terminals. E no cover MCP servers, and e no cover Omnigent supervisor process itself. MCP server wey you start go run outside the box with your permissions. Na this gap make the stronger pattern still be one throwaway machine per agent. running coding agents in a disposable VM talk about this. The other half of the work na credentials, and keeping secrets out of an agent's reach go become harder, not easier, when six sub-agents share one host.
Spend thresholds na policies wey you declare for the same file:
policies:
budget:
type: function
handler: omnigent.policies.builtins.cost.cost_budget
factory_params:
max_cost_usd: 5.00
ask_thresholds_usd: [1.00, 3.00]The cost_budget policy dey ask for approval at the soft thresholds and dey block all models at max_cost_usd by default. If you set a non-empty expensive_models list, the limit become downgrade gate: models wey no dey that list fit continue spending. Run wey plan with one vendor, implement with another, and review with a third dey spend money for three places at once. So set the thresholds before the first unattended run, instead of waiting until the first invoice. The built-ins also include max_tool_calls_per_session and ask_on_os_tools, wey ask for approval before file and shell operations. Our notes on keeping AI agent costs under control on a VPS apply directly here, and dem apply even more because parallel sub-agents dey multiply the burn rate.
Repository dey move how fast?
Di data wey dey behind dis chart
[
{
"version": "v0.2.0",
"released": "2026-06-19",
"interval": 3
},
{
"version": "v0.3.0",
"released": "2026-06-27",
"interval": 8
},
{
"version": "v0.4.0",
"released": "2026-07-03",
"interval": 6
},
{
"version": "v0.5.0",
"released": "2026-07-10",
"interval": 7
},
{
"version": "v0.5.1",
"released": "2026-07-10",
"interval": 0
},
{
"version": "v0.6.0",
"released": "2026-07-21",
"interval": 11
},
{
"version": "v0.7.0",
"released": "2026-07-27",
"interval": 6
}
]Na the release dates wey the project publish for e own releases page, as dem read am on 3 August 2026. 7 tagged releases land between 2026-06-19 and 2026-07-27, and the longest gap between any two na 11 days. v0.5.1 release happen the same day as the release before am. The first two PyPI releases, 0.1.0 on 13 June 2026 and 0.1.1 on 16 June 2026, no get entry for the GitHub releases page, so dem no dey inside the chart; the three days wey show for v0.2.0 count from 0.1.1.
Two of those releases change commands wey guides don already document. v0.7.0 remove omni server start and use omni server --background instead. v0.6.0 rename the omnigent[memory] extra to omnigent[hindsight], but e keep memory as compatibility alias wey still work through 0.10.0. From 0.11.0, wey release on 25 August 2026, the alias don disappear. So install line wey you copy from June write-up no go pull Hindsight client again. Na this one show why you need --version for your install command and tag for your git clone; no be style preference.
The speed no slow down since that reading. By 29 October 2026, PyPI list 0.20.0, wey release on 28 October, thirteen minor releases after 0.7.0, and the project release notes flag breaking changes for several of dem. Two matter for anything wey you copy from this page: v0.12.0 make bare omni start the background host instead of the Polly agent (omni run now launch the default agent), and v0.13.0 require tmux 3.3 or newer for managed terminals.
Wetin I no go trust am with yet
As of 27 October 2026, the repository get about 11k stars, 1.8k forks and roughly 910 open issues, while the first public release don pass just over four months. Stars show interest, but interest no mean maturity. The project talk say na alpha, and the release history above show say e really mean alpha.
- I no go run am for host wey hold production credentials, because the sandbox no cover MCP servers or the supervisor.
- I no go leave run unattended without
cost_budgetpolicy, because three vendors fit bill at the same time; with non-emptyexpensive_modelslist, spending fit continue on cheaper models. - I no go expose the server on public IP address without setting
OMNIGENT_AUTH_ENABLEDand putting TLS in front of am. - I no go treat the agent YAML as stable across minor versions yet, so pin the version and read the release notes before you upgrade.
One more thing you need know before e surprise you: v0.6.0 add anonymised usage telemetry, the README talk say dem collect am by default, and the project document am, including how to opt out, for dedicated telemetry page. Read that page and decide deliberately if the machine dey handle client work.
Wetin Omnigent genuinely good at today na the work wey dem build am for. You get three or four agent CLIs, you already dey pay for dem, and you want one to write while another one review am. This one dey work now, for one machine, with real sandboxing on Linux. Treat everything beyond that as promising but unfinished.
FAQ
Omnigent na agent, or na thing wey dey run agents?
E dey run agents. Omnigent na meta-harness: e dey start vendor CLIs wey you don already install, like Claude Code, Codex or OpenCode, give each one work, then supervise the results for one session. E no bring any model of its own. Na this make e different from framework, where you write Python against library and your own program become the agent.
I need install Claude Code and Codex before Omnigent go useful?
You need at least one vendor CLI wey you don install and log in to, because Omnigent dey drive those programs instead of replacing dem. For the shipped Polly example, you need two or more from different vendors. Polly rule be say different vendor from the implementer must always do the review, so if na only one CLI dey available, no second vendor dey to send the diff to.
How I fit install specific Omnigent version instead of the latest one?
Pass --version through the install script with sh -s --, as shown for sh -s -- --version 0.7.0. If you no use -s --, sh go take --version as its own option and the installer no go run: Bash, for example, go print its own version and exit. If uv don already dey available, uv tool install --force --python 3.12 "omnigent==0.7.0" go do the same work. The git tag na v0.7.0, while the PyPI version string na 0.7.0.
Omnibox sandbox enough to run agents unattended?
E strong for wetin e cover, and e clear about wetin e no cover. For Linux, e dey use bubblewrap plus seccomp, so kernel dey enforce file and network limits, and agent no fit opt out. The documentation talk say e dey apply to sys_os_* tool calls and terminals, but e no cover MCP servers or the Omnigent supervisor process. Therefore, MCP server go run with your normal permissions. Na why disposable virtual machine for each agent still give stronger isolation for unattended work.
How much memory Omnigent server need for VPS?
The project's deploy notes report about 275 MB RSS when idle, and its Fly.io configuration allocate 1 GB machine. The idle measurement cover the server alone, not its sub-agents. Sub-agents fit share the parent's working directory and environment, including its sandbox through os_env: inherit; separate working copies no dey guaranteed. Polly-style runs dey use parallel git worktrees, so size RAM and disk based on the number of agents you plan to run at the same time, not based on the server.