How to self-host Agentlas OS for VPS with Ollama
Install Agentlas OS v1.2.0 for Linux VPS with pinned command, find where state dey, connect Ollama, and know why idle hub costs disk space, not RAM.
Wetin Agentlas OS really be
Agentlas OS na open source agent runtime wey keep specialist agents for disk as packages and build temporary orchestrator for each task. You fit self-host am by installing am inside your own user account for Linux VPS. The pinned v1.2.0 command-line installation wey dey below no be service: the installer no enable daemon or open listening port. The current project also document separate desktop app.
This last difference determine everything else for this page. Most multi-agent systems dey run supervisor process wey remain active and hold the agents. Agentlas turn this arrangement around: the specialists na files wey dey stored, while the orchestrator dey exist only when task dey run. The practical result be say idle hub go use your disk, but e no go use memory.
The project call its open core Hephaestus, and na this name you go see for commands, paths, and environment variables. The repository na agentlas-ai/Agentlas-OS, e get Apache-2.0 license, and dem write am mostly for Python.
How early this project be, honestly
As recorded on 13 October 2026, the repository dey about four months old, with roughly 1,580 stars and 110 forks. That one still young for something wey you wan use for real work.
The release cadence matter pass the age. Version v1.1.103 publish on 8 August 2026, and v1.2.0 land on 12 August 2026. That mean over one hundred tagged releases for the 1.1 series, with some days getting several releases, all published by automation. The 1.2 series continue with the same pace: e don reach v1.2.62 by 10 October 2026. Project wey dey move that fast fit change behaviour under you between Tuesday and Thursday.
So pin the release. The installer dey read environment variable for this, and the whole guide below dey use am. If you install this kind project without pinning the version, you fit get whichever release dem tag most recently when dem last edit the script.
Wetin you need for the VPS
The requirements no plenty because idle specialists na files, no be processes.
- A Linux VPS. Ubuntu 24.04 na good baseline. The installer dey detect the operating system with
uname -sand e dey use non-macOS branch for Linux, so headless box dey supported. curl,tarandgitfor the box, plus Python 3.9 or newer. The pinned installer dey use am to verify the bundled embedding model.- Outbound HTTPS to
raw.githubusercontent.comandgithub.comfor normal installation. The installer dey download release archive and check its SHA-256. The current standalone installer dey rejectHEPHAESTUS_SOURCE_DIRwithstandalone_local_source_unverified; offline installation through that variable no longer dey supported. - A host harness, wey be the coding agent wey actually dey talk to a model. Claude Code, Codex, opencode, goose and Hermes all dey supported as adapters.
You no need root. The installer dey write only under your home directory, including ~/.local/bin. When one step fail, like path wey no writable, e go warn and continue with the remaining steps, then e go exit with non-zero status at the end if anything fail. If you still dey choose the box itself, run a coding agent on a VPS dey cover the base image and access setup wey this setup dey build on top.
Install the pinned release
The upstream README dey document one line wey dey pipe script from main straight into bash. Download the v1.2.0 installer and read am first. E dey write into every agent harness wey e find, so e worth make you use ten seconds check am.
curl -fsSL -o install-all-runtimes.sh \
https://raw.githubusercontent.com/agentlas-ai/Agentlas-OS/v1.2.0/scripts/install-all-runtimes.sh
less install-all-runtimes.sh
HEPHAESTUS_REF=v1.2.0 bash install-all-runtimes.shHEPHAESTUS_REF na the pin. As of 8 November 2026, the script for main dey use release_ref="${HEPHAESTUS_REF:-os-v1.2.89}", and the releases page mark Agentlas OS 1.2.89, wey dem release today under os-v1.2.89, as Latest. The current installer need os-v tag and e dey reject HEPHAESTUS_REF=v1.2.0 with standalone_channel_tag_invalid. The historical installer above accept that pin.
The script URL above pin both the installer and the runtime payload to v1.2.0. Current standalone installs dey use ~/.agentlas/os-runtime and the ~/.agentlas/os-bin/agentlas-os launcher, and dem no longer install global wrappers or register host plugins and MCP permanently by default. Those defaults different from the historical installation wey dey below.
The pin cover the install, but e no cover wetin happen afterwards. For v1.2.0, hephaestus commands dey start background update check. When newer release dey available, e install am and repoint ~/.agentlas/runtime/current without asking. The installer disable this for its own run with HEPHAESTUS_AUTO_UPDATE=0, and na the same variable you go use to keep this historical pin. E go enter your shell profile for the next step. Current standalone automatic updates dey off by default.
If the run succeed, e go print the paths wey e write, including these lines when ~/.local/bin no dey inside PATH:
Installed runner: /home/you/.agentlas/runtime/current/bin/hephaestus
Installed shell commands in /home/you/.local/bin (add ~/.local/bin to PATH to use them)That PATH warning na the line wey people dey skip. For fresh Ubuntu box, ~/.local/bin often no dey inside PATH, so every hep-* command go fail with command not found even though the install succeed. Add am to ~/.bashrc, disable the historical background updater at the same time, then confirm:
echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc
echo 'export HEPHAESTUS_AUTO_UPDATE=0' >> ~/.bashrc
source ~/.bashrc
hep-global statushep-global na the original spelling of hephaestus global, and status show whether each harness instruction file get the managed router block. That router dey off by default, so expect say e go report none. If the command run at all, your PATH correct.
Where the state dey live
Everything na file inside your home directory, so backup and migration dey simple.
~/.agentlas/runtime/1.2.0/dey hold the runtime itself (the installer dey remove the leadingvfrom the tag), while~/.agentlas/runtime/current/na symlink to the active version. Two pinned versions fit dey side by side.~/.local/bin/dey hold the shell wrappers, includinghephaestus,hep-build,hep-network,hep-search,hep-storm,hep-cloudandhep-upload.~/.agentlas/networking/memory/dey hold the durable memory:playbook-registry.json,playbook-candidates.jsonlandmemory-events.jsonl.~/.agentlas/networking/hub-agents/<slug>/memory/experience.sqlitedey hold experience for each agent, scoped by owner.<project>/.agentlas/ontology-runtime.sqlitedey hold state for each project, so e dey move with the repository instead of the box.~/.cache/agentlas/pythondey hold the Python cache for Linux. macOS dey use another path, and na the installer choose that branch withuname.
The v1.2.0 memory documentation talk clearly say secrets, raw credentials and full transcripts must not enter any memory scope. Credential values dey stay for gitignored local files, while memory records only names and paths. Back up ~/.agentlas and your project .agentlas directories, and you fit rebuild everything for a new VPS.
One thing no dey remain for the box: for the pinned v1.2.0 release, hep-build dey send anonymous build counters by default, without prompts, paths or agent content. Set AGENTLAS_TELEMETRY=0 or DO_NOT_TRACK=1, or put both "telemetry": false and "telemetrySetBy": "user" inside ~/.agentlas/networking/config.json, to turn am off. Current standalone telemetry dey off by default.
Wich model backend e fit point to
Na this detail dey change how you understand the whole setup: Agentlas no dey call model API. Na the host harness dey do am.
The architecture document describe runtime adapters wey translate one core into each harness, and the README talk say adapter own provider authentication, while caller provide im own model and API access. Agentlas provide two surfaces wey harness fit pick up: an AgentSkills file and an MCP (model context protocol) server wey dey communicate through stdio. So the question “which models Agentlas support” really mean “which models your harness support”, and the answer na any model wey Claude Code, Codex, opencode, goose or Hermes fit reach.
Registering the MCP server dey look like this for Codex-style TOML config:
[mcp_servers.hephaestus-network]
command = "~/.agentlas/runtime/current/bin/hephaestus"
args = ["mcp", "serve"]The same server go register inside ~/.cursor/mcp.json, ~/.config/goose/config.yaml and the other harness configs automatically during install. If you dey connect several of these for one box, run MCP servers for a VPS explain the stdio and process model with more detail.
Point am go self-hosted Ollama endpoint
Because harness dey control model connection, to point Agentlas at local models mean say you go point your harness go Ollama. Ollama add launch subcommand for v0.15 exactly for this purpose, and e still dey ship for v0.32.9 as of 11 August 2026. E configure existing harness against local models without any environment variable to set:
ollama pull qwen3-coder:30b
ollama launch opencodeUse claude, codex or droid instead of opencode, depending on the harness wey you install. Then route one request through the local runtime:
~/.agentlas/runtime/current/bin/hephaestus route "summarise the failing tests" --runtime ollamaIf route work, e go return JSON decision wey name the agent or team wey e select, together with receipt_id. If e no return anything useful, check the context length. The v1.2.0 local-models documentation ask for model wey get at least 64k context for sessions wey routing heavy, and e name qwen3-coder, gemma3 and deepseek-r1 as examples. Ollama own guidance for coding tools na the same 64k minimum. Routing decisions carry agent inventory inside the prompt, so model wey get 8k or 32k context fit truncate the inventory and choose badly.
One limitation wey the tagline no go tell you. Ollama, Gemma and DeepSeek no get plugin or command system of their own, so /agentlas slash commands no dey there. Harness wey you launch against Ollama still provide its commands, skills and plugins; but raw Ollama API loop use the MCP server or hephaestus route command instead. For raw API loop, this na real reduction in available features.
Wetin one hub of idle specialists dey cost for RAM
Nothing. Na the complete answer, and you fit prove am instead of just trust am.
Borrowed hub specialists dey arrive as package artifacts, no be processes. One specialist na an agent.md plus one .agentlas/ directory wey contain JSON: routing-card.json for triggers and capabilities, memory-map.json for write boundaries, mode-map.json for whether e dey run alone or as team. Hephaestus Network documentation describe runner wey dey launch workers for task, and nothing inside am open listening port or ask you to enable service. Between tasks, check am yourself:
pgrep -af hephaestus
systemctl --user list-units --type=service | grep -i agentlas
du -sh ~/.agentlasOnce command-triggered workers don exit, first two commands go print nothing for idle box, because specialists no dey resident. The pinned v1.2.0 fit leave short-lived, detached update and build-telemetry workers running after command return. HEPHAESTUS_AUTO_UPDATE=0 dey disable im updater; telemetry need im own opt-out wey dem describe above. Current standalone update checks and telemetry dey off by default. The third one print the only cost wey parked hub impose on you, and na disk. E dey grow based on the number of specialists wey you keep plus the bundled embedding model wey the runtime ship.
So, the memory question na completely question about the burst, and the burst na your harness plus your model backend. If the harness dey talk to hosted API, the resident cost na the harness process, and im memory use depend on the harness and workload. If you self-host the weights, na the weights be the bill:
Di data wey dey behind dis chart
[
{
"label": "Hosted API model",
"weights_gb": 0
},
{
"label": "gemma3:4b",
"weights_gb": 3.4
},
{
"label": "gemma3:12b",
"weights_gb": 8.2
},
{
"label": "gemma3:27b",
"weights_gb": 17
},
{
"label": "qwen3-coder:30b",
"weights_gb": 19
}
]Those na the published download sizes from Ollama's model library, no be measurements from benchmark run, and KV cache for 64k context dey add on top every figure above zero. The model wey Agentlas docs name first, qwen3-coder:30b, need 19 GB of weights before context, and even the 27B Gemma variant dey ask for 17 GB. Compare those numbers, Agentlas layer itself no show for the budget.
How e take compare with running one harness
Run one harness against a hosted API, and na your VPS go carry one process. Add Agentlas, and e go carry the same harness, plus files and task-scoped runtime processes. The orchestrator no be another long-lived program. Na bigger prompt wey e assemble from the packages for disk, then e discard am.
The cost wey change na context, no be memory. Orchestrator wey pull several specialist cards and their routing metadata go use more tokens for each task than bare harness. For hosted API, na money this one go cost, no be RAM. For local weights, na time e go cost, because longer prompt mean longer prefill for CPU or busier GPU.
Na why sizing advice for box like this dey follow model decision, no be agent framework. How to size RAM and CPU for coding agent VPS explain am well, and the conclusion still hold here: choose plan based on backend wey you plan to run, then add a couple of gigabytes headroom for the harness. If you prefer the always-on supervisor design for comparison, the Omnigent multi-agent harness keep its coordinator resident. Na opposite trade-off be this, and e go show directly for idle memory.
Failure modes and the strings wey you go see
hep-build: command not found right after clean install. The pinned installer write to ~/.local/bin, wey no dey inside some fresh shells' PATH. E talk am near the end of the output, but the warning scroll pass. Add the export wey dey show above to ~/.bashrc.
Behaviour change after you rebuild the box. You no set HEPHAESTUS_REF, so the installer use whatever tag e script name that day by default. Pin am, and record the pin beside your other version numbers.
Behaviour change without rebuild. For the pinned v1.2.0, the background updater fit install newer release and move ~/.agentlas/runtime/current go there. Check wetin current dey point to, and set HEPHAESTUS_AUTO_UPDATE=0 for the environment wey your harness dey run inside, no be only interactive shell. Current standalone automatic updates dey off by default.
Routing dey pick wrong specialist for local model. Check whether the model context window too small for the agent inventory. Move go model wey get 64k or more, and set Ollama context length to match. Ollama default to 4k below 24 GiB VRAM, 32k from 24 to below 48 GiB, and 256k for 48 GiB or more, so the smaller configurations need explicit increase.
ollama launch no dey recognised. The subcommand enter Ollama v0.15. Older packages from distribution repository don come before am, so install current Ollama.
The install dey write inside harnesses wey you no expect. The pinned script detect and configure every harness wey e find, writing to ~/.claude/, ~/.codex/, ~/.gemini/, ~/.cursor/ and more. E also edit config files wey already dey exist, like ~/.codex/config.toml, while e keep backup beside the original. For shared build box, read the script before you run am, and know which of those directories matter to you.
Make you run this one now?
Project wey still young and dey get automated releases, sometimes several for one day, no suppose carry production workload yet. The architecture really dey interesting, the license na Apache-2.0, and all the state dey inside files wey you fit back up. But removing am no as clean as that description suggest: the v1.2.0 README no document any uninstall process, the pinned installer writes for different places inside your home directory and edits harness configs wey already dey there, while hephaestus global remove only removes the optional router block. These facts make am cheap to test for disposable box, but expensive to depend on.
For now, better approach na to pin v1.2.0 with its tagged installer, set HEPHAESTUS_AUTO_UPDATE=0 so this historical pin go remain, run am for box wey you fit rebuild, keep ~/.agentlas inside your backups, and read the changelog again before you move the pin. If you want wider survey of other options for this space and how mature each one be, this roundup of self-hosted AI agents na better place to start, while self-hosting a Hermes agent on a VPS explain one of the harnesses wey Agentlas dey adapt to.
FAQ
Agentlas OS dey run as server for my VPS?
The pinned v1.2.0 command-line installation no enable daemon or open listening port. The installer write runtime inside ~/.agentlas/runtime/ and command wrappers inside ~/.local/bin. Hephaestus Network dey run when task launch am, instead of as background service. After command-triggered workers don exit, pgrep -af hephaestus no print anything for idle machine. The pinned release fit launch short-lived update and build-telemetry workers: HEPHAESTUS_AUTO_UPDATE=0 disable updater, while AGENTLAS_TELEMETRY=0 or DO_NOT_TRACK=1 disable telemetry. Current standalone update checks and telemetry dey off by default. Self-hosting here mean say the code and state dey your machine; e no mean say service dey listen.
How much RAM idle specialists hub dey use?
None, because idle specialists no be processes. Specialist na agent.md file plus .agentlas/ directory wey hold routing-card.json, memory-map.json and similar metadata. So parked hub only cost disk space. Use du -sh ~/.agentlas to measure am. Tasks and short-lived background workers dey use memory; parked specialists no dey use memory.
Which models I fit use, and I fit point am to my own Ollama?
Agentlas no dey call model APIs by itself. Host harness dey manage credentials and connection, so supported models na any model wey your harness support. For local weights, run ollama launch opencode, and substitute claude, codex or droid. This go configure the harness against your Ollama server without environment variables. Use model wey get at least 64k context, like qwen3-coder or gemma3, because routing prompts carry the agent inventory and smaller windows fit truncate dem badly.
Which version I suppose install, and why pinning matter here?
Install v1.2.0, the tagged release wey current on 12 August 2026, by fetching the installer from the v1.2.0 tag and setting HEPHAESTUS_REF=v1.2.0 as shown above. As of 8 November 2026, the script default for main na release_ref="${HEPHAESTUS_REF:-os-v1.2.89}", and Agentlas OS 1.2.89 na the latest release, published today under os-v1.2.89. That current installer accept only os-v version tags and reject the old v1.2.0 pin. Pinning matter pass normal because the project publish over a hundred releases for its 1.1 series, with several releases on some days. So, if you rebuild weeks later without pinning, you no go get the system wey you test. Also set HEPHAESTUS_AUTO_UPDATE=0 for the historical v1.2.0 runtime, because otherwise its updater go install newer releases for background and the pin go quietly stop holding. Current standalone automatic updates dey off by default.