> For the complete documentation index, see [llms.txt](https://doc.autoreach.tech/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://doc.autoreach.tech/settings-and-configuration/ai-models.md). # AI Model Configuration AutoReach runs AI across many tasks (finding and scoring leads, writing and replying to messages, enrichment, and more). How that AI is powered, and whether you can choose models, depends on your billing mode. ## How AI is billed You have two options: * **Included credits (default).** AI runs on AutoReach's models, billed in credits. You start with **free credits** and can top up anytime from the **Credits** page. There are no API keys to configure. On credits, each task runs on a tuned default model, so model selection is managed for you. * **Bring your own key (optional).** Add your own **OpenAI, Anthropic, or DeepSeek** key in **Settings > AI & Models** for unlimited usage billed directly by your provider with no markup. With your own key, you can also choose the primary and fallback model for each category below. > Email finding uses your own **Findymail** API key, which is separate from the above and only needed for email discovery. ## Model Categories AutoReach uses AI across multiple categories. Each has independent primary and fallback model settings: | Category | Purpose | | ------------------ | ----------------------------------------------------------------- | | Content Writing | DMs, replies, first messages, templates, warmup, copilot | | Buyer Scoring | Fit/intent/timing analysis and lead qualification | | Pain Inference | Infers each lead's pain from their full profile before scoring | | Keyword Generation | X + LinkedIn search keyword generation | | Classification | Objection detection, negative content, deal signals, ICP matching | | Profile Finding | Email, LinkedIn, X profile, and website discovery | | Web Enrichment | Company research, news, social profiles, insights | | Lookalike Search | Finding influencers similar to target profiles | | Research & Copilot | Follow-up research, copilot web search, topic research | | Lead Relevance | Post and tweet relevance analysis using reasoning | | Call Briefs | Pre-call meeting preparation documents | ## Per-User Overrides > Choosing models per category requires your own API key (the bring-your-own-key option above). On included credits, every category runs on a tuned default and the model picker is not used. The cost guidance in the rest of this page applies to bring-your-own-key usage, where you pay your provider directly; on credits you are billed in credits and see your usage on the **Credits** page. Each category can be customized with a primary and fallback model selection. The system validates your choices against available models and checks that the required API key is present. Invalid or missing configurations fall back to defaults. ## How Fallbacks Work If the primary model fails (timeout, rate limit, API error), AutoReach automatically switches to the fallback model for that operation. This provides resilience without manual intervention. You do not need to do anything when a fallback occurs- the system handles it automatically and the output is still delivered. The only visible effect is that your cost reflects the blended rate of both models rather than just the primary. If both the primary and fallback models fail (rare, but possible during provider outages), the operation is retried later. ## When to Change Model Settings **Good reasons to change models:** * You want to reduce costs by using a cheaper model for high-volume categories like scoring or classification * You want higher quality DMs or posts and are willing to pay more for a premium content writing model * You have API keys for only one provider and need all categories to use that provider's models **When to leave defaults alone:** * If you are just getting started, the defaults are a solid starting point * If your current output quality and costs are acceptable, there is no need to experiment * If you are unsure which models to use, the defaults are well-tested ## Cost Implications Model pricing varies significantly. As a rough guide: * **Nano-class models** are the cheapest option - ideal for scoring, classification, and keyword generation where volume is high * **Mid-tier models** (Sonnet-class) offer a good balance and work well as fallback models * **Premium models** (Opus-class) deliver the highest quality but cost considerably more per token - reserve these for content writing if budget is a concern Since scoring and classification run on every lead while content writing only runs when sending messages, moving scoring to a cheaper model produces larger savings than optimizing content writing. ## Recommended Configurations **Budget-conscious setup:** Use nano models for scoring, classification, and keyword generation. Use a mid-tier model for content writing. This minimizes per-lead costs while keeping message quality reasonable. **Quality-focused setup:** Use a premium model for content writing (DMs, posts, replies). Keep scoring and classification on nano or mid-tier models - these tasks do not benefit as much from premium models. This gives you the best message quality without overspending on scoring. ## Settings UI The AI Configuration section in Settings shows all categories with primary and fallback dropdowns. Changes take effect immediately and are validated on save. ## Dynamic Cost Impact Your model selections directly affect pipeline costs. The cost estimation system reads your actual model configuration and calculates blended pricing based on observed fallback rates. See [**Cost Estimation**](/settings-and-configuration/cost-estimation.md) for details. ## Next Steps * [**Pipeline Cost Estimation**](/settings-and-configuration/cost-estimation.md): How model choices affect per-lead costs --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://doc.autoreach.tech/settings-and-configuration/ai-models.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.