What is Lalabase?
Lalabase is project management for software agencies, freelancers and small software companies that work for clients. Each client is an organisation with its own projects, tickets on a list or board, time tracking with several billable sessions per ticket, notes and documents, a docs folder synced from GitHub or GitLab, and meetings stored as transcript plus notes. Clients log in with rights set per area and per project, and their accounts do not take a paid seat.
The AI layer is off until an admin switches it on. It then answers questions across tickets, comments, documents, released meetings and the indexed source code, using pgvector search in PostgreSQL, and lists the sources under each answer. The provider is your choice: OpenAI, Anthropic, Google Gemini through Vertex AI in an EU region, the EU presets for Mistral, IONOS and OVHcloud, or any OpenAI-compatible endpoint. An MCP server (0.15, 19 tools) and a REST API let Claude, Cursor or your own agent read and write tickets and time.
It is a Ruby on Rails 8.1 and PostgreSQL app under AGPL-3.0-only, made by gboyke solutions UG in Rosengarten, Germany. The public repo was created on 11 October 2026 and has about 101 stars, 6 forks and one contributor; the README says the product has been developed privately since October 2024 and used by paying customers since September 2025. Releases are monthly and date-numbered, the latest being 2026.11.0 on 31 October 2026.
Is Lalabase worth it?
Lalabase's distinctive part is not the ticket board but the answer under it. Ask how many hours went into a client's checkout work, or what was promised in last month's call, and the assistant answers from the tickets, comments, documents, released meetings and indexed source code of that project, with clickable sources underneath. According to the feature pages, project permissions carry into the chat, so no answer draws on an area the asker may not read, and the assistant only reads: it creates no tickets and changes no status.
The AI design favours control over convenience. AI is off by default, and when it is switched on you pick the provider: OpenAI, Anthropic, Gemini on Vertex AI pinned to an EU region (the docs reject Google AI Studio and the global endpoint for data residency), EU presets for Mistral, IONOS and OVHcloud, or any OpenAI-compatible endpoint. The hosted Cloud AI plan uses Google, OpenAI and Mistral, with an allowance whose size is not published. One detail to check before planning a local model: the 2026.10.1 changelog says a custom endpoint entered in the app must resolve to a public address, which sits awkwardly with the README's promise of a model in your own network.
Meetings are a deliberate half-feature. Lalabase records nothing and transcribes nothing; you paste a transcript from whatever tool ran the call, and the AI writes a summary, works out who spoke and suggests tickets that only exist once someone accepts them. Each meeting stays out of search and chat until it is released. That is safer for client trust than a bot joining calls, and it means a separate transcription service remains in your stack and your data processing agreements.
The hosting story is unusually plain. The cloud runs at Hetzner in Germany with a DPA, and the privacy policy lists OpenAI and Anthropic in the US, Google Cloud in Ireland and Mistral, IONOS and OVH in the EU as AI recipients, depending on the package you choose, so German hosting does not by itself keep prompts in the EU. The security page says there is no ISO 27001 certificate, content is not end-to-end encrypted, and only the founder can reach customer data. Under AGPL-3.0-only an agency can run it for clients unmodified at no cost; patch it and give clients access, and you must offer those clients your source. The README also says unsolicited pull requests are not accepted.
Maturity is the open question. The code shows a long private history, more than 2,300 commits and about 970 pull requests by the README's count, and the changelogs read like a careful operator's: upgrade notes, signed images and fixes such as access tokens no longer written to the cache in plain text. But the public repo has one contributor and about 101 stars, and the company page names one founder. Against OpenProject it admits it has no Gantt charts or portfolios. For a ten-person agency already on Jira or Redmine, the real comparison is whether one founder's roadmap beats a larger tool's ecosystem; the AGPL source is the insurance if it does not.
Where it earns its place
A twelve-person web agency in Germany billing clients by the hour
Answer client questions about hours and status without digging through three tools
Run the hosted Cloud AI plan, give each client a free read-only login to tasks, and set notes to no access so internal comments stay internal. Twelve staff at €13 each is €156 a month net, or €128 per person a year paid up front.
A platform engineer at an agency with a strict no-US-AI policy from its clients
Self-host project management with AI that stays within EU providers
Deploy with docker compose on a 2 vCPU, 4 GB server, then set the IONOS or Mistral preset, or Gemini on Vertex AI in europe-west3, as the only provider. Licence cost is zero; changing the embedding size later means a full console re-index.
A lead developer who works in Claude Code
Let the coding agent pick up tickets, log time and look up project knowledge
Create a personal access token and connect the MCP server, which in 2026.11.0 has 19 tools for tickets, comments, time, the knowledge base, people and indexed code; the separate Lalabase CLI can hand a ticket to Claude Code from the terminal.
Before you commit
- Setup
- Clone the repo, check out the latest release tag, fill in .env and run docker compose up -d, then bin/rails lalabase:setup. The README asks for 2 vCPU, 4 GB RAM, 20 GB disk and Docker 24 or later; compose starts the app, a worker, PostgreSQL with pgvector, Valkey and Caddy for TLS.
- Integrations
- GitHub and GitLab for docs folders and code indexing, an MCP server with 19 tools, a REST API (1.8.0) with idempotency keys, and the separate Lalabase CLI for macOS and Linux. No repository hosting and no Gantt view.
- Data and privacy
- Hosted at Hetzner in Germany with a DPA; logs deleted within 7 days; 30 days to export after cancelling. AI is off by default and receives only the context a feature needs, sent to the provider you pick, which may be in the US. Credentials and backups are encrypted; content is not. No ISO 27001.
- Cost at scale
- Self-hosted: free, plus a small server and your AI provider's bill. Hosted: €10 per staff member a month with your own key, €13 with Lalabase's AI, 20% less paid yearly; client logins are free. New companies can apply for 50% off for six months.
- Limits
- Public since October 2026 with one contributor; monthly releases with upgrade steps to read; no built-in transcription; rights per area rather than per ticket; hosted AI allowance not published; embedding size changes need a console re-index.
What’s good
- Cited answers across tickets, comments, docs, released meetings and code
- You choose the AI provider, including EU-only presets and Vertex AI in Frankfurt
- Client accounts are free; rights set per area and per project
- Careful changelogs with upgrade notes and signed multi-arch Docker images
What’s not
- Public repo three weeks old with a single contributor
- No built-in transcription; transcripts must be pasted in
- No ISO 27001, no Gantt or portfolios; rights are per area, not per ticket
- Hosted AI allowance size not published; outside pull requests not accepted
Best for
Software agencies of a few to a few dozen people in the EU who want tickets, time, docs and client access in one self-hostable tool with an assistant they control.
Lalabase alternatives
Lalabase: common questions
- Is Lalabase worth it?
- Lalabase is a careful, privacy-minded agency tool whose cited answers across tickets, meetings and code are worth a look, but the public project is three weeks old and rests on one developer. We rate it 3.6 out of 5.
- How much does Lalabase cost?
- AGPL-3.0-only, free to self-host. Hosted in Germany (prices net of VAT, business customers only): Cloud €10/person/month with your own AI key or no AI; Cloud AI €13/person/month with a monthly AI allowance per organisation; 20% off yearly (€8 and €10.67). Client accounts are free and unlimited.
- What are the downsides of Lalabase?
- Public repo three weeks old with a single contributor; no built-in transcription; transcripts must be pasted in; no ISO 27001, no Gantt or portfolios; rights are per area, not per ticket; hosted AI allowance size not published; outside pull requests not accepted.
- Who is Lalabase best for?
- Software agencies of a few to a few dozen people in the EU who want tickets, time, docs and client access in one self-hostable tool with an assistant they control.
- How much does Lalabase cost for a team or in production?
- Self-hosted: free, plus a small server and your AI provider's bill. Hosted: €10 per staff member a month with your own key, €13 with Lalabase's AI, 20% less paid yearly; client logins are free. New companies can apply for 50% off for six months.
- Is Lalabase safe for company data?
- Hosted at Hetzner in Germany with a DPA; logs deleted within 7 days; 30 days to export after cancelling. AI is off by default and receives only the context a feature needs, sent to the provider you pick, which may be in the US. Credentials and backups are encrypted; content is not. No ISO 27001.
- What are the best Lalabase alternatives?
- The closest alternatives we have reviewed are Macro.
Answers checked Oct 31, 2026.