# Agentic Search Source: https://docs.mira.day/en/docs/agent-sourcing/agentic-search > Describe the person you want in plain language. Mira searches across multiple data sources, digs in step by step, and returns candidates with the reason each one matches. Agentic Search is how Mira finds people. You don't write boolean strings or juggle keywords — **describe who you're looking for in plain language, and Mira works like a seasoned recruiter: it searches across multiple data sources, follows the leads step by step, and tells you why each candidate matches.** *** What it solves [#what-it-solves] Traditional candidate search has two built-in limits: * **One database.** The right people are scattered across professional networks, code communities, research platforms, and company sites — a single database is bound to miss some. * **One pass.** A single keyword match tends to return people with good-looking résumés, not people who are actually a good fit. Agentic Search is different on both counts: it pulls from **multiple data sources** and advances in **multiple steps** — deciding where to look next from what the last step surfaced — until it finds the right people, including passive candidates who aren't actively looking. How it works [#how-it-works] 1. **You describe the need** — in everyday language: the role, the must-haves, and the nice-to-haves. 2. **Mira searches** — across multiple data sources, following leads (companies, tech stacks, projects, public work) and cross-checking them. 3. **You get explainable results** — a candidate list where every person comes with *why they match*. 4. **You refine** — if the direction is off, just say so or add a constraint, and Mira adjusts. Core features [#core-features] * **Cross-source recall** — searches multiple professional talent sources and the open web at once, for broader coverage. * **Multi-step deep dig** — not a one-shot match; it follows leads and cross-verifies, so it can reach passive candidates. * **Explainable matching** — every candidate carries a reason, so results are judgeable and trustworthy. This is powered by [MRE](./understanding-ai-matching). * **On-demand contact** — once the right people are found, reveal email or phone as needed (each reveal uses credits). Describe the role well [#describe-the-role-well] The Agent understands natural language — but more context gives it more to reason about. A weak prompt and a strong one return very different shortlists: | Weak | Strong | Why it matters | | ------------------------- | ----------------------------------------------------------------------------------------------------- | ------------------------------------------------------ | | "Find me a developer" | "Senior backend engineer, 5+ years Python, has built microservices at scale, preferably from fintech" | More context = more signals to match against | | "Marketing person in NYC" | "B2B SaaS marketing manager in NYC, has run product launches, comfortable with data-driven campaigns" | Implicit criteria trigger reasoning-based matching | | "We need someone good" | "Someone who's taken a product from 0 to 1, ideally at a Series A–B startup" | Mira reads "0 to 1" as early-stage building experience | You can lean on implicit criteria too — the kind you'd tell a recruiter out loud ("scrappy", "can operate independently", "built a team from scratch"). Mira reasons about what they mean rather than matching them as keywords. Quick start [#quick-start] 1. **Describe the person in plain language**, covering as much as you can: * Role / level * Must-haves (the non-negotiables) * Nice-to-haves * Constraints like location, industry, or company type 2. **Wait while Mira searches** — it shows the steps as it works. 3. **Review the list and each candidate's match reason**; reveal contact details as needed. 4. **Not quite right?** Follow up or adjust the criteria, and let it keep refining. **Example prompt:** "Find 3 senior backend engineers who've built real-time risk control at fintech companies, ideally with Go and Kafka, based in Shanghai." Mira searches across sources for people who fit, digs into their projects and background to cross-check, and returns a shortlist with a match reason for each — then you can say "find a few more with team-management experience" to narrow it down. Tips for better results [#tips-for-better-results] * **Separate must-haves from nice-to-haves.** Tell Mira what's required vs. what's a bonus — matching gets sharper. * **Give the background.** "This is an early-stage startup that needs someone who can operate independently" helps it judge fit. * **Use follow-ups.** When the direction is off, add or correct a constraint instead of starting over. * **Let complex roles run as a team.** For cross-functional or heavily-constrained roles, hand it to an [Agent Team](./agent-teams) to work in parallel. * **Save your go-to plays.** Turn a search workflow you repeat into a [Skill](./skills) so the team can reuse it. Under the hood [#under-the-hood] A few things power Agentic Search, in plain terms: * **Multi-source recall + fusion ranking.** Mira queries several data sources at once, then merges, de-duplicates, and re-ranks the results (reciprocal rank fusion) so no single source's bias dominates. * **Iterative reasoning loop.** Rather than one query, it reasons → searches → reads the results → reasons again, choosing each next step from what it just found. * **Layered pipeline.** Discovery, contact enrichment, and profile enrichment run as separate layers, going deeper only as needed. * **Explainable matching.** [MRE](./understanding-ai-matching) evaluates each candidate against your requirements and writes the reason you see. > In academic terms, combining many heterogeneous sources and solving step by step across domains is **multi-source retrieval** plus **multi-hop reasoning**. Related [#related] * [Review your Shortlist](./review-your-shortlist) — read and export the candidates you get back. * [AI candidate matching](./understanding-ai-matching) — how MRE evaluates fit. * [Agent Teams](./agent-teams) — put several specialized agents on a complex search. Ready to try Mira? [Join the waitlist](https://www.mira.day/join-waitlist) for early access.