Private AI Lead Assistant
A locally run AI CRM tool used to extract structured lead information from customer conversations. All analysis is performed locally, and data is not uploaded to cloud servers.
Timeline (last 90 days)
Private AI Lead Assistant is an AI CRM tool that runs entirely on the user's local machine, used to extract structured lead information from customer conversations.
Private AI Lead Assistant's tech stack includes React 19, Tailwind, Express, Postgres and TypeScript, and there is no external AI API call anywhere in the codebase.
Private AI Lead Assistant's workflow includes a mandatory human review step, and no extraction result is saved until the user confirms it.
Through an evaluation of 29 real conversations, the developer found that when the conversation does not mention urgency, the model defaults to labeling the priority as medium.
Private AI Lead Assistant uses deterministic sampling parameters for extraction tasks: temperature=0, seed=42, numCtx=8192, numPredict=512.
Private AI Lead Assistant uses temperature=0.7 when generating follow-up messages, so as to produce different email content each time.
Private AI Lead Assistant uses Zod to validate the JSON output by the LLM, and when validation fails it returns a 502 error with detailed error information.
Private AI Lead Assistant uses a regex detector to flag prompt-injection-shaped content before the text reaches the model, and combines it with the system prompt and review UI warnings to form layered protection.
The developer explicitly states that the 1.5B model sometimes gives wrong answers confidently, so human review is used as the final gate.
Private AI Lead Assistant's target users are freelancers and small agencies of 2 to 10 people, and the 1.5B model size was deliberately chosen so it can run on a laptop CPU and keep the tool free.
1 more timeline events
All Facts (11)
Private AI Lead Assistant is an AI CRM tool that runs entirely on the user's local machine, used to extract structured lead information from customer conversations.
50%UnverifiedPrivate AI Lead Assistant's tech stack includes React 19, Tailwind, Express, Postgres and TypeScript, and there is no external AI API call anywhere in the codebase.
50%UnverifiedPrivate AI Lead Assistant's workflow includes a mandatory human review step, and no extraction result is saved until the user confirms it.
50%UnverifiedThrough an evaluation of 29 real conversations, the developer found that when the conversation does not mention urgency, the model defaults to labeling the priority as medium.
50%UnverifiedPrivate AI Lead Assistant uses deterministic sampling parameters for extraction tasks: temperature=0, seed=42, numCtx=8192, numPredict=512.
50%UnverifiedPrivate AI Lead Assistant uses temperature=0.7 when generating follow-up messages, so as to produce different email content each time.
50%UnverifiedPrivate AI Lead Assistant uses Zod to validate the JSON output by the LLM, and when validation fails it returns a 502 error with detailed error information.
50%UnverifiedPrivate AI Lead Assistant uses a regex detector to flag prompt-injection-shaped content before the text reaches the model, and combines it with the system prompt and review UI warnings to form layered protection.
50%UnverifiedThe developer explicitly states that the 1.5B model sometimes gives wrong answers confidently, so human review is used as the final gate.
50%UnverifiedPrivate AI Lead Assistant's target users are freelancers and small agencies of 2 to 10 people, and the 1.5B model size was deliberately chosen so it can run on a laptop CPU and keep the tool free.
50%UnverifiedFor Private AI Lead Assistant, open source is a precondition for the product to exist, because the input contains sensitive data such as client budgets and NDA plans, and local inference keeps conversation records on disk.
50%