Office docs to Markdown in milliseconds, plus YC's fleet-of-agents harness

The Lens

anydoc converts Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV, and PDF files into clean Markdown, with a median conversion under 5 milliseconds per document. The reason to care is feeding documents to an LLM without maintaining a pile of format-specific parsers. Every format parses into one shared document model and renders through a single Markdown serializer, so a table from a spreadsheet and a table from a Word file come out the same shape. MIT licensed. The core is Rust with bindings for Node, Python, and the browser through WebAssembly, so install is a package manager one-liner in whichever language you already use. There is no server to run and no state to manage. Format detection reads file contents rather than trusting the extension, which means a mislabeled upload does not blow up the pipeline. Headings, lists, tables, footnotes, and cross-references survive the trip. Free for everything local. Firecrawl sells a hosted product, Parse, which is the same conversion plus OCR models. That is the honest line: if your documents are scanned images rather than real files, you need OCR, and OCR is the paid side. The catch is that "clean Markdown" is carrying weight. Heavily designed PDFs, multi-column layouts, and deeply nested tables remain hard, and no converter has fully solved them. Test it against your ugliest real document before you build a pipeline on top of it.

The Lens

Most CRMs bolt an AI assistant onto a form. Comp AI's CRM inverts that: the agent runs on its own schedule and the CRM is simply where it writes things down. It researches contacts, enriches records, and books follow-ups without waiting for you to click anything, and every contact carries an Agent tab showing the reasoning and what it plans to do next. MIT licensed, and one of the fastest-moving projects we track right now. Self-hosting means three deployments and a Postgres: a Next.js app, a NestJS API, and the agent itself. You need Bun and Docker, plus OAuth credentials from Google or Microsoft. Redis is optional. It runs with no external API keys at all, with LinkedIn, web search, and company data as optional enrichment sources you can add later. Free, all of it, with no paid tier and no hosted upsell. The agent sandbox runs deny-all egress with no direct database access, which is a sane default for software making autonomous decisions about your customer records. Solo operators and small teams get the whole product. The catch is age. This is a young project moving very fast, and an autonomous agent writing to your customer data is a lot of trust to extend to code this new. The evidence-based design helps, since tools are built to report observations rather than guesses, and nothing about a person is invented. Still, point it at a copy of your data before you point it at the real thing.

The Lens

SSO (single sign-on), multi-factor auth, user directories, all in one place. The open source version is extremely capable. You get SAML, OAuth2/OIDC, LDAP, SCIM, and proxy authentication. The admin UI is clean and modern. You can set up enrollment flows, password recovery, and MFA without touching code. For a homelab or small company, this replaces paid services like Okta or Auth0. Authentik Enterprise adds features like long-term support, AI-powered risk assessment, and enterprise support starting at $5/user/mo with a 100-user minimum ($500/mo floor). Solo/homelab: the open source version is fantastic. Small teams: same. Growing teams needing vendor support: Enterprise makes sense at $5/user/mo. Large orgs: serious Okta alternative at a fraction of the cost. The catch: initial setup has a learning curve. Authentik uses a flow-based system for authentication logic that's powerful but not intuitive at first. Plan to spend a weekend getting it right. Docker Compose is the recommended deployment; you'll run Postgres, Redis, and the Authentik containers.

The Lens

qm is Y Combinator's own open source answer to running a fleet of AI agents across a team. It's a "multiplayer agent harness for work": every person gets an isolated workspace with personal and shared scopes, you drive it from Slack or a web UI, and it can run background and scheduled agent jobs. This is YC Labs code, not a fan project trading on the name, the release is announced on ycombinator.com and served from qm.ycombinator.com. The interesting part is the multiplayer framing. Instead of each developer babysitting one agent in a terminal, qm treats agents as shared, always-on coworkers with their own scopes and cron schedules. For a team already leaning on coding agents, it's a free way to run them coordinated instead of siloed. Solo use works too, but the design clearly targets groups. The catch is that it's brand new. The commit history is short and the stars arrived fast, which means the code is early and production-readiness is unproven. The pedigree is real and the idea is sharp, but pilot it on non-critical work before you wire it into how your team actually ships.