Skip to main content

Use an Installed Coding Agent as a Chat Model

If you already pay for a coding agent and it is signed in on the machine running Libre WebUI, you can talk to it from a normal chat instead of adding a separate API key. Libre WebUI detects the agent's command-line interface and offers it in the model selector like any other model.

What gets detected​

On each request for the model list, Libre WebUI looks on the server's PATH for these commands:

ModelCommandProvided by
Claude CodeclaudeAnthropic
CodexcodexOpenAI
Kirokiro-cliAWS
Muse CodemuseMeta
OpenCodeopencodeSST
PipiMario Zechner

Anything found appears under an Agents group in the model selector. Nothing is installed for you, and no configuration file is required — if the command runs in your server's shell, it shows up.

Settings → Chat → Defaults also identifies these entries as Agent, including Strands when your account may use it. Plugin models keep their provider name, and only Ollama models carry the Ollama label. Disabling Ollama does not hide configured agents or plugins. The model information card shows the selected entry's display name and provider, with size, family, and format only when supplied by its catalog.

Choosing a model, not just a CLI​

Each CLI can expose several entries in the Agents group:

  • Claude Code offers its signed-in default plus Sonnet, Opus, Haiku, and explicit Sonnet 5.5 (claude-sonnet-5-5), Opus 5.5 (claude-opus-5-5), and Haiku 5.5 (claude-haiku-5-5) choices. Sonnet 5.5 requires Claude Code 2.1.284 or later, Opus 5.5 requires Claude Code 2.1.280 or later, and Haiku 5.5 requires Claude Code 2.1.293 or later, all with access through the signed-in account; run claude update to update the CLI. The existing Sonnet, Opus, and Haiku aliases continue to follow the CLI's current model mapping.
  • Codex offers its configured default plus the documented ChatGPT sign-in family: GPT-6 Astra, GPT-6.1 Sol (gpt-6.1-sol), GPT-6 Sol, GPT-6 Luna, GPT-5.6 Sol, Terra, Luna, GPT-5.5, and GPT-5.3 Codex Spark. Availability depends on the CLI's sign-in and account access.
  • Kiro lists the models its signed-in account can use (from kiro-cli chat --list-models) beside its configured default. Each turn runs non-interactively on the CLI's V3 agent engine with its built-in default agent (vibe) pinned, so a custom chat.defaultAgent does not apply. Requires Kiro CLI 2.27 or later.
  • Muse Code offers its configured default plus Muse Spark 1.3 (muse-spark-1.3), Muse Spark 1.3 Contributor (muse-spark-1.3-contributor), and Muse Wonder 1.0 (muse-wonder-1.0). It signs in with muse login (a Meta account) or a Meta Model API key from META_API_KEY or muse auth set. Contributor models are billed at a lower rate because Meta may train on what you send them.
  • Pi runs with the model configured in the CLI itself.
  • OpenCode lists the models of every provider it is authenticated with (from opencode models), and always requires an explicit choice — its CLI-level default can point at a local server that is not reachable from the Libre WebUI host.

The embedded Strands engine also appears in the Agents group when your account may use it. It is not a CLI: it runs inside the backend and drives your Ollama and provider plugin models. The plain Strands entry (strands) uses your default chat model. The other entries (strands:<route>) list one per available model, such as Strands · gpt-5.6-sol (Codex (ChatGPT)), and pin both the model and its provider for that conversation; another available provider cannot replace it during an outage. Listing choices does not start the engine.

The same Strands entries work as the task model under Settings → Chat → Defaults for automatic titles and thinking summaries. Use current running model uses the conversation's saved Strands selection. These short text requests call its underlying provider model directly; they do not run the agent's tools or create an agent session. Other CLI profiles require an Ollama, plugin, or Strands task model for these auxiliary requests.

Pi runs each turn stateless (--no-session), with local tools disabled and a neutral system prompt, so replies are not colored by — and chats never touch — the personal Pi configuration of the server's operating-system user.

Kiro is the opposite on both counts: it runs with the server user's Kiro configuration — the built-in default agent is pinned, but permission rules (~/.kiro/settings/permissions.yaml), MCP servers, and the configured default model still apply — and every turn is saved as a new session in that user's Kiro session store (kiro-cli chat --list-sessions shows it, with the recent chat transcript inside). There is no flag to turn that off, so chat content also lives in the backend user's home directory, outside the Libre WebUI database. Kiro's tools stay enabled: whatever those permission rules trust — by default, reads and read-only commands such as date — runs without asking, and any tool call that needs approval is refused, because a chat turn cannot answer the prompt.

Muse Code reads the prompt from a private file and runs headless with approvals off (--approval-mode never), since a chat turn cannot answer an approval prompt. Its own sandbox stays on, and each turn gets a scratch workspace that is deleted afterwards, so its file writes cannot land in the server user's home. Turns are not saved in Muse's session log (--no-session-log), but the server user's Muse settings, rules, and default model apply.

Using it​

  1. Sign in as an administrator.
  2. Start a chat and open the model selector.
  3. Pick an entry under Agents.
  4. Send messages as usual. Replies stream back token by token.

The conversation is stored like any other chat, so you can leave and come back to it, rename it, and search it from the command palette. Each turn sends a transcript of the recent conversation to the agent and streams its answer back.

Who can use it, and what it can reach​

This feature is administrator-only, and the reason matters. The agent runs as the same operating-system user as the Libre WebUI backend, on the host — not inside a Work container. It therefore inherits that user's agent credentials and whatever access those agents normally have, including the ability to read files and run commands on the server if it decides to.

Treat enabling this as equivalent to granting the agent shell access to the machine. Because of that, the feature ships disabled: an administrator must turn it on under Settings → User Management → Access & policies → Agent CLI models. The setting is persisted and takes effect immediately, without a restart.

Changing this control refreshes the model catalogue immediately; the Agents category contains direct CLI entries, while Plugin Models contains provider API integrations such as Codex (ChatGPT). Strands is listed in the same category but follows its own Strands engine access setting, not this toggle, so it can appear while Agent CLI models are off. Its model discovery cannot block discovery of installed CLI agents.

Existing installations inherit their previous saved Agents decision until an independent CLI setting is saved.

To pin the decision at the deployment level regardless of the runtime toggle, set the environment variable either way (this also locks the toggle in the interface):

AGENT_CLI_MODELS_ENABLED=false   # or true

For a model that can act on files but stays inside a sandbox, use Work instead: it runs tool-capable models in a locked-down container with an isolated workspace.

The same agents in Work​

Work offers the same agents with the same model choices and the same logins, but runs them inside the task's sandbox instead of on the host, behind an egress proxy that keeps the real tokens out of the sandbox. They have their own access mode, Agent CLIs in Work, and do not need the switch above. See Agent CLIs in Work.

Usage tracking​

Open Provider Usage (/usage) and find Agents near the top. Claude Code, Codex, Kiro, Muse Code, OpenCode, Pi, and Strands activity have explicit entries, including a clear message when the selected period has no recorded calls.

CLI calls made through LWUI record their outcome, duration, and reported token usage. Each invocation creates one usage event; repeated usage snapshots and per-step reports do not create duplicate events or token counts. Missing token counters remain unreported. Stopping a response records cancellation, and an unsuccessful CLI exit does not become a success merely because it produced partial text. Usage from CLI calls outside LWUI is not imported.

Strands calls are attributed to the Strands agent. See Agent usage for the breakdown and limits.

Configuration​

VariableDefaultPurpose
AGENT_CLI_MODELS_ENABLEDunsetPin the feature true/false; unset defers to the admin toggle (off by default)
AGENT_CLI_TIMEOUT_MS600000How long a single agent turn may run before it is killed

Troubleshooting​

No Agents group appears. Confirm the feature is enabled under Settings → User Management → Access & policies → Agent CLI models (it is off by default) and that you are signed in as an administrator, then check that the command is on the PATH of the process running the backend — not just your interactive shell. A service manager, Docker container, or desktop launcher often starts with a much smaller PATH than a login terminal.

The reply fails immediately. Run the same command by hand as the server user (claude -p "hello", codex exec "hello", muse exec "hello", or echo hello | kiro-cli chat --no-interactive --output-format stream-json --agent-engine v3 --agent vibe). Kiro CLI 2.27 or later is required; an older CLI fails here first, where its error is visible — in chat the same failure only shows as Kiro exited unsuccessfully. Most failures are the agent asking for a login that has expired, or a rate limit on the underlying subscription. OpenCode in particular reports an expired provider login only in its own logs; re-run opencode auth login as the server user. Kiro signs in through kiro-cli login; an expired session fails the same way. Muse Code says missing meta credentials until muse login has run as the server user.

Replies stop partway. A long turn may have hit AGENT_CLI_TIMEOUT_MS. Raise it, or break the request into smaller steps.