update
May 26, 2026
By Teun
Harbor v0.4.19 adds launcher for local coding tools
Harbor v0.4.19 adds a new `harbor launch` command for starting pre-configured local coding tools and pointing them at inference backends. The release also adds web search and URL-reading support through Harbor Boost, plus compatibility and hardening updates.
Harbor v0.4.19 introduces a new harbor launch command that starts pre-configured local agentic coding tools on the host and connects them to inference backends running in Harbor. The release notes say the command picks the first reachable running backend by default, but also supports explicit backend and model selection.
The examples show harbor launch codex as the simplest form. Users can also set the backend and model before the tool name, such as harbor launch --backend ollama --model qwen3.5:4b codex, harbor launch --backend ollama --model qwen3.5:4b mi, harbor launch --backend llamacpp --model Qwen3.5-4B opencode, and harbor launch --backend ollama --model qwen3.5:4b copilot -p "explain this repo".
Harbor says the new command can also route requests through Harbor Boost when users add --web. In that mode, the launched tool can use web search and URL-reading tools, which is intended to extend the agent beyond local code and model access.
The release notes also call out support for Claude Code using Harbor’s Ollama backend through Ollama’s Anthropic-compatible API. An example shown is harbor launch --backend ollama --model qwen3.5:4b claude -p "explain this repo".
Harbor also added a --config option that prints the computed launch configuration without starting the tool. The examples include harbor launch --config opencode and harbor launch --config openclaw, which suggests the launcher can be used to inspect how Harbor will wire up a tool before actually running it.
Beyond the launcher, the changelog mentions an Anthropic-compatible Messages API and OpenAI Responses API compatibility layer for Boost. It also references a service for starting host tool adapters, ik_llama.cpp service support, and a Boost workflows and tools module.
The release notes say Boost also gained adaptive thinking, annotations, rate limit forwarding, and SSE keep-alive support in the compatibility layers. Harbor describes additional hardening work around streaming fixes, an SDK compatibility audit, and request validation.
The changelog closes with new guides, docs, and screenshots, along with a link to the full v0.4.18 to v0.4.19 comparison.
