issue of integration between Ruflo and Ollama

[ilanbrightspeed-hash]

am trying to cconfigure ruflo with ollama , but still the following ::

Version: 3.67.0
OS: Windows
Provider: Ollama
Model: gemma3:4b

Symptoms:

• providers test => PASS
• AI workers => enabled
• daemon => running
• workflow => running
• Execute step => pending forever
• task => in_progress forever
• ollama ps => always empty
• no workflow/task logs generated

key log line
Claude Code not found - AI workers will run in local fallback mode

[Sagargupta16]

Nothing in that setup makes a model call, so Ollama never receives a request. Three separate things are going on, all in the 3.67.0 source.

The daemon line is expected and has nothing to do with Ollama. AI workers run only through the Claude Code CLI: services/headless-worker-executor.ts probes claude --version (isAvailable(), line 713) and runs each worker as spawn('claude', ['--print', '--output-format', 'json', ...]) (line 1423). No provider setting reaches that path. Without claude the workers fall back to local mode, where audit does file-presence checks and no analysis (worker-daemon.ts line 1805). "AI workers enabled" plus "local fallback mode" is the designed outcome on a machine without Claude Code.

workflow run only writes a record. workflow_run (mcp-tools/workflow-tools.ts, line 135) creates the steps as pending, sets the workflow to running with agentsSpawned: 0, saves and returns; the default template has one stage, named Execute, which is the step you are watching. Nothing in the CLI walks it afterwards. Execution is a different tool, workflow_execute (line 320), which has no workflow subcommand (run, validate, list, status, stop, template) and whose task steps need an agentId. Tasks are the same kind of record: task_create stores pending, task_assign flips it to in_progress (task-tools.ts line 511), and only task_complete changes it again. in_progress means assigned, not running.

The one place a model is called is agent_execute (mcp-tools/agent-execute-core.ts), and it does support a local Ollama, but only from an MCP client (Claude Code, or any client pointed at ruflo mcp start); no CLI subcommand calls it. For that path the self-hosted branch needs an explicit base URL (lines 228 to 242):

set OLLAMA_BASE_URL=http://localhost:11434
set OLLAMA_DEFAULT_MODEL=gemma3:4b
set RUFLO_PROVIDER=ollama

or the persisted form, ruflo providers configure -p ollama -e http://localhost:11434 -m gemma3:4b. providers test passes without any of that because it falls back to localhost:11434 on its own (providers.ts line 385); execution does not. OLLAMA_URL from the user guide is not read anywhere in the CLI; OLLAMA_BASE_URL is.

Read from the source, not run on Windows.