Your company’s information is everywhere. Files, calls, meetings, email, sheets, scans, threads, code. Runlog AI understands it all and gets things done. One memory your people and your tools both read from, every answer traced to its source.

Runlog reads 97 formats across 6 kinds of file: Documents (13, such as pdf, docx, pptx, md, txt); Tables (11, such as xlsx, csv, parquet, tsv, ods); Images (11, such as png, jpg, heic, tiff, svg); Audio (9, such as mp3, wav, m4a, flac); Video (9, such as mp4, mov, webm, mkv); Code (44, such as py, ts, go, java, sql).

Every answer comes with provenance. Asked: Ray retires in March. What knowledge leaves with him? Runlog answered: GTC-04 escalation is excluded from the 7% IRR test under a 2024 side letter, and that rule exists only in his comment on row 5; no executed agreement in the data room states it, and every option analysis applies it. Sources: 2024 side letter, Deliverables tracker, Ray Alvarez. It also produced a memo, GTC-04 escalation, reconciled (2 pages · 3 sources · drafted from the project). The exchange shows inbuilt provenance for each answer, context retention across sessions, deep synthesis across multimodal data.

One AI that does it all Because the better answer is also the cheaper one. SWE-Explore-Bench, hitfile (precision): Runlog AI 93.1% at $2 tokens, Claude Code 66.7% at $63 tokens, Codex 64.9% at $50 tokens. BEAM, 10M tokens, answers right: Runlog AI 59.8%, Hindsight 58.3%, mem0 48.6%. The subscriptions that stop: coding agent, task agent, agent memory, vector search.

Start on web or install on your computer. Install: curl -fsSL https://get.runlogai.com/install.sh | sh Runlog runs as Web Portal (Work from anywhere), Desktop App with CLI (Working directories and codebase), Runlog API (We love devs), Enterprise (SSO, audit export, your terms). Dashboard: https://atlas.runlogai.com. Docs: https://www.runlogai.com/docs. Book a meeting: https://calendly.com/runlogai/30min.

Models change. Company memory compounds. Every claim resolves to the span it was read from, and stays true when the source moves. We will never sell your prompts for a discount. Your company already has the information. Runlog turns it into memory. Encrypted at rest and in transit, per-project access control, hard-delete that cascades. SOC 2 Type 2 in progress. Pricing: https://www.runlogai.com/pricing. Demo: https://www.runlogai.com/demo.

What is Runlog, and who is it for?

Runlog is company memory. It reads every document, call, sheet, thread and repository a company has, holds them as one map of how people, contracts, deals and code connect, and answers from that map with a source on every line. It is built for teams whose value lives in their records: consulting and advisory firms, construction and field-services firms, small companies selling into a niche market, and developers pointing their own tools at a codebase.

How is Runlog different from ChatGPT, Claude, or Copilot?

Chat assistants answer with whatever context you paste into the conversation, and forget it after. Runlog maintains a persistent, source-linked map of your business that updates as your data changes, and runs recurring workflows on top of it. Assistants sit on top of Runlog well; they replace it poorly.

How is it different from RAG or a vector database?

A RAG pipeline retrieves the closest passage to each question and forgets the rest; a vector database matches similar text. Runlog resolves what each document means, keeps the map current when a source changes, and treats people, contracts and deals as real things with a history. Every answer cites the document, page and row it came from, and a citation to something that does not exist is dropped before anyone reads it.

Does it work with Claude Code, Cursor, or my own product?

Yes. One command binds a folder or a repository, and Claude Code, Cursor, ChatGPT or your own product answer from the whole of it through the API and the MIT-licensed Memory MCP server, with the file and line behind every claim. On SWE-Explore-Bench, Runlog names the right file 0.931 of the time against 0.667 for Claude Code, using about 16 thousand tokens instead of roughly 630 thousand. The desktop app watches folders on your own machine, so what changed today is part of tomorrow morning’s answer.

What can Runlog read?

97 formats across six kinds of file: documents, tables, images, audio, video and code. PDFs and scans, spreadsheets, recorded calls and meetings, email and chat threads, web pages, and repositories in forty-four languages. Each format has its own reading path, and they all land in one shape.

What does Runlog cost?

Individual is $79 a month and carries a $19 weekly allowance of metered work; Individual Max is $249. Team is $1,500 a month flat for unlimited people on one pooled allowance, and Team Max is $2,500. Annual is ten months for twelve. Past the weekly allowance, prepaid credits draw at twice incurred cost. Reading a project costs nothing, and seats cost nothing. Enterprise is quoted, never published.

Where does our data live?

In our managed offering, in per-project isolated namespaces with hard-delete that cascades through every derived asset. Encrypted in transit and at rest. We are working toward SOC 2 Type 2 and have an on-prem architecture specified but not yet shipped, so if either is a hard requirement today, say so early and we will tell you honestly where we stand.

How do we start?

Sign in, point Runlog at a folder, and read what it comes back with. The first session reads the folder, asks only what it cannot infer, and returns a cited page. Nothing needs a call first. Enterprise terms, white-labelling and on-prem are the conversation that does.