Relaticle

About Relaticle

Relaticle is an open-source CRM built around AI agents rather than having AI added as an afterthought. It launched with two distinct trust levels for AI writes: external MCP clients authenticate over OAuth and write directly through 37 tools, while the in-app assistant generates proposals that require record-by-record human approval. The entire system is self-hostable under the AGPL license, with support for local model inference through Ollama.

Review

Relaticle takes a cautious approach to letting AI modify customer data, which sets it apart from CRMs that simply bolt on text generation. The creator spent roughly two years building this solo, and the codebase reflects a developer who ran into real-world problems with AI agents and rebuilt the chat client to handle failures like rate-limited retries and stale stream handlers. What ships today is a functional CRM with agent capabilities, though some features discussed publicly remain in the idea stage.

Key Features

• 37 first-party MCP tools available to any MCP-compatible client over OAuth, with workspace custom fields appearing in each agent's schema automatically
• Approval-gated in-app assistant that shows the exact change as a proposal card and requires human approval or skip for each record before any data is modified
• Self-hosting under AGPL with documented support for Ollama, keeping customer data local when running model inference
• Rebuilt chat client that handles failure states: drafts survive page reloads, failed sends remain visible, and answers link back to referenced records
• Flat per-workspace pricing for the cloud version, with a free-forever self-hosted tier

Pricing and Value

Self-hosting Relaticle is free under the AGPL license. The cloud version uses flat per-workspace pricing rather than charging per seat. Specific cloud pricing figures are not yet defined in the available materials. A 50% discount on the first year was mentioned at launch.

Pros

• Two-tier trust model gives external agents direct write access while keeping in-app changes behind explicit approval, matching different use cases
• Self-hostable with local inference means teams can avoid sending customer data to third-party model providers
• Custom fields propagate into agent schemas without manual configuration, reducing setup work
• Chat client addresses concrete failure modes like conversation corruption from rate-limited retries
• Flat per-workspace pricing avoids the cost scaling issues of per-seat models for larger teams

Cons

• No undo trail for approved writes currently exists; once you approve a change, reversing it requires manual correction
• Policy-based auto-approval for the in-app assistant is not available-every write stays explicit, which creates friction as trust builds
• The tool is not well suited for teams that want AI to silently update records based on learned behavior or external triggers without human review

Relaticle fits teams that want AI to do real work on CRM data but aren't comfortable with silent writes. The per-record approval model makes sense for sales teams handling sensitive client information, while the MCP compatibility serves developers who already use Claude Desktop or Cursor and want those tools to interact with their CRM. Organizations that need AI to operate autonomously on their database will find the current approval requirements limiting.

Open 'Relaticle' Website