RagLeapragleap-core v0.9.0 was published on 2026-10-08. Its release notes call it the biggest update since v0.7.0: RagLeap is now an open-source AI-employee platform you can install in minutes, with a dashboard, a command-line tool and a choice of embedding providers. This post summarises what is in it, and what the release notes say is still missing. The full notes are on the release page and on GitHub.
The AI Office dashboard
The dashboard lives at /office. It has an overview, approvals, employees by department, a task board, agent runs, an activity log and a Settings tab. In Settings you pick the AI provider and models, paste your keys (they are stored encrypted and never shown again) and test the connection. A value saved in the dashboard wins over .env, and the page shows where each value comes from. By design, the dashboard is reachable locally or through an SSH tunnel.
An installer and the ragleap command
The installer asks which AI to use (Gemini, fully local Ollama with no API key, or skip), generates your API key and secrets, starts everything and prints a one-click link to the dashboard. An existing .env is never modified. For unattended installs you can set RAGLEAP_PROVIDER to gemini, ollama or skip, together with GEMINI_API_KEY.
Afterwards the ragleap command controls the stack: launch (start; with --ollama it also runs the local AI service and switches to it), stop, status, logs, update (which backs up the database first) and key, which prints your dashboard link. The install guide is at docs.ragleap.com/install.
Embeddings are no longer Gemini-only
Embeddings still use Gemini by default, but EMBEDDING_PROVIDER now switches them to Ollama, OpenAI or Mistral, and the code also lists Together, OpenRouter, Qwen, Zhipu and a custom OpenAI-compatible endpoint. For providers without a built-in default you set the embedding model and its dimensions yourself. The vector size is checked at startup, and tables that already hold embeddings are never changed. See the configuration page and the providers page.
Agents, tasks and approvals
- Agents: an act-observe agent loop, sandboxed code and shell execution, read-only page fetching and an MCP client.
- Tasks and triggers: task tickets, and scheduled triggers with cron expressions and time zones.
- Approval inbox: approve or reject actions in the dashboard or over the API.
Safety
- Approval modes (off, semi and full), and roles in sensitive domains always need approval.
- A taint rule that applies after web content has been read, a sandbox without network access, and SSRF-safe page fetching.
- API-key protection on all data routes, and no exception text returned to callers.
Honest limits
The release notes list these themselves:
- The installer is tested offline and has not yet had a real clean-machine run.
- Local models on a CPU are slower and less accurate than Gemini.
- Public access, login throttling, an interactive browser and more channels are not built yet.
Upgrading
Run ragleap update, or git pull and docker compose up --build -d. The database changes are additive.
Read more
The documentation is built from this release. For an outside view of what RagLeap Core offers, including its 46 built-in roles, see RagLeap Core: 46 built-in AI employee roles. The source is on GitHub.