• Browser Preview — Open generated HTML files in default browser with javascript execution warnings. • Playground Import — Load ZIP or Markdown skills directly with automatic formatting validation. • Simplified Playground — Save setups directly as single-agent workflows or ZIP skills. • Multiple Server Connections — Configure and switch between multiple remote servers with optional API keys. • Some bug fixes and improvements.

• Browser Preview — Open generated HTML files in default browser with javascript execution warnings. • Playground Import — Load ZIP or Markdown skills directly with automatic formatting validation. • Simplified Playground — Save setups directly as single-agent workflows or ZIP skills. • Multiple Server Connections — Configure and switch between multiple remote servers with optional API keys. • Some bug fixes and improvements.

GPU/NPU (if available) can be used for embedded models and some bug fixes

• Agent Orchestration — Build multi-agent workflows with sequential or conditional routing. • Agent Visual Builder — Interactive 2D flowchart canvas with zoom, pan, and subagent/routing configuration. • Multiple Server Connections — Configure and switch between multiple remote servers with optional API keys. • Modern UI design style option. • Toggle native tool calling for Ollama models in LLM settings. • Safe tool call mode for Ollama models using GBNF grammar-constrained decoding. • Some bug fixes and improvements.

some bug fixes and improvements

Agent Orchestration — Build multi-agent workflows with sequential or conditional routing. Agent Visual Builder — Interactive 2D flowchart canvas with zoom, pan, and subagent/routing configuration. Modern UI design style option.

Agent Orchestration — Build multi-agent workflows with sequential or conditional routing. Agent Visual Builder — Interactive 2D flowchart canvas with zoom, pan, and subagent/routing configuration. Modern UI design style option.

Some bug fixes and improvements.

Some bug fixes and improvements.

Modern UI design style option. • Toggle native tool calling for Ollama models in LLM settings. • Safe tool call mode for Ollama models using GBNF grammar-constrained decoding. • Some bug fixes and improvements. • Release-notes. • Sub-prompt editor — split any prompt into sequential steps with ++#++; use ${tool_result} to pipe the previous step's tool output into the next.

Modern UI design style option. • Toggle native tool calling for Ollama models in LLM settings. • Safe tool call mode for Ollama models using GBNF grammar-constrained decoding. • Some bug fixes and improvements. • Release-notes. • Sub-prompt editor — split any prompt into sequential steps with ++#++; use ${tool_result} to pipe the previous step's tool output into the next.

some generic improvements

improved website-index inbuilt tool

New feature: Toggle native tool calling for Ollama models in LLM settings.

New feature: Toggle native tool calling for Ollama models in LLM settings.

New feature: Toggle native tool calling for Ollama models in LLM settings.

New feature: Toggle native tool calling for Ollama models in LLM settings.

• Embedded on-device models — download GGUF models and run inference on your device; no API key or internet needed. Discover, download and remove models directly from Settings → LLM. • Local + server improvements — conditional email behavior is now aligned across both modes, with custom SFTP support for agent output delivery • System prompt optimization — when tools are selected, tool skills are automatically injected into the system prompt; compact SLM hints used for small/local models; regenerate or customize skills per MCP tool in Settings → Tool Skills.

• New feature: "Stop after tool call" switch per sub-prompt to optimize embedded model performance. • Some bug fixes and improvements. • Release-notes. • Sub-prompt editor — split any prompt into sequential steps with ++#++; use ${tool_result} to pipe the previous step's tool output into the next. • Tool skills — auto-generated, editable LLM usage hints per MCP tool; build skills from any MCP server (GitHub, JS bridge, custom) in one tap. • Server Mode full support — connect to TealKit on your home server with complete local ↔ server parity for agent behavior and settings; agents run 24/7 without the app open • Embedded on-device models — download GGUF models and run inference on your device; no API key or internet needed. Discover, download and remove models directly from Settings → LLM. • Local + server improvements — conditional email behavior is now aligned across both modes, with custom SFTP support for agent output delivery • System prompt optimization — when tools are selected, tool skills are automatically injected into the system prompt; compact SLM hints used for small/local models; regenerate or customize skills per MCP tool in Settings → Tool Skills.

some smaller improvements - detect non standard tool calls from embedded models

some smaller improvements - detect non standard tool calls from embedded models

What's new in this version: - App is completely free from v1.2.1 — all features available to everyone, no trial, no purchase required. - Server Mode — run TealKit as a headless server on a Linux machine, Jetson Nano, Mac Mini or Raspberry Pi; connect from the desktop app remotely and schedule agents to run 24/7 without the app being open; auto-generated API key, Docker-ready with Dockerfile included; server image coming next - Embedded on-device models — download GGUF models and run inference on your Mac; no API key or internet required. Go to Settings → On-device models to get started. - Playground setups — save the current tool selection, system prompt and initial prompt as a named setup; reload any saved setup in one tap to restart the playground with the same configuration - SLM mode — new checkbox in LLM 1 and LLM 2 settings; generates a short directive system prompt that forces immediate tool calls (ideal for Ollama, Phi, Mistral-small and other local models) - LLM 1/2 selector in Playground — choose which configured LLM to use per session; LLM 2 option appears only when configured - Ollama model browser — load available models from your server (/api/tags) directly in LLM settings; autocomplete dropdown + model validation on Test Connection - LLM 2 advanced settings parity — Max Tool Output Size and Token Warning Threshold now configurable for LLM 2 (same as LLM 1) - System prompt optimization — tool skills are automatically injected into the system prompt when tools are selected; compact SLM hints for small/local models; customize per MCP tool in Settings → Tool Skills.

Embedded model configurable in global llm settings

bug fixes

• Sub-prompt editor — split any prompt into sequential steps with ++#++; use ${tool_result} to pipe the previous step's tool output into the next. No Pro required. • Tool skills — auto-generated, editable LLM usage hints per MCP tool; build skills from any MCP server (GitHub, JS bridge, custom) in one tap. • 31-day free trial — all Pro features available during the trial period. • Server Mode full support — connect to TealKit on your home server with complete local ↔ server parity for agent behavior and settings; agents run 24/7 without the app open • Embedded on-device models — download GGUF models and run inference on your device; no API key or internet needed. Go to Settings → On-device models. • Local + server improvements — conditional email behavior is now aligned across both modes, with custom SFTP support for agent output delivery • Tool list export — tap the training icon on any MCP chip or server tile to export tools as OpenAI functions, Anthropic tools, Markdown, or JSONL. Copy to clipboard or save to file.