Rust·Latest release v0.3.4 · 2026-11-09

KYB

Git-backed infrastructure knowledge base for engineering teams and AI agents.

Latest releasev0.3.42026-11-09
Last commit7 days ago2026-11-10
crates.io0.3.487 downloads
Articles3on this blog
Recent releases
All 14 releases →
Recent commits
  • ca774a5 Merge pull request #62 from alex09x/docs/cite-0.3.4 2026-11-10
  • 78d49aa docs: cite the 0.3.4 Zenodo record 2026-11-10
  • 08837e3 Merge pull request #61 from alex09x/feat/confident-english-only 2026-11-09
  • d65cdba confident: every field of a record is English 2026-11-09

Status from GitHub and crates.io as of 2026-11-16.

What it is

KYB (Know Your Business) is shared, versioned memory about your own infrastructure: which servers exist and what runs on them, how services are wired and deployed, which decisions were made and why - plus incident reports and tasks.

Coding agents forget everything between sessions. Several agents on several machines rediscover the same infrastructure again and again, and every lesson from an incident disappears when the context window closes. KYB is the memory that survives: agents query it before reasoning about a system and write back what they verified.

  • Git is the canon: one Markdown file per entry, one commit per change - a commit sha is a version id.
  • Hybrid search: BM25 fused with multilingual vector search (multilingual-e5-small, int8 ONNX, ~5 ms on CPU).
  • Every historical version stays searchable - ask what the knowledge said before it changed.
  • Writes that look like tokens, private keys or passwords are rejected.

How it works

Git is the database

Entries live under knowledge/, incidents/ and tasks/ as Markdown with YAML frontmatter. History, diff and rollback come from Git; the canon can be read and edited with any text editor.

The index is a cache

Tantivy is rebuilt from Git on every start and covers the latest version of every key and every historical version. Closing an incident or task archives it: it leaves the working tree but stays in the default search, marked archived.

Search finds meaning across languages

Lexical and semantic candidates are combined with reciprocal rank fusion, so a question in one language lands on an entry written in another while exact technical terms still rank first. Without a model on disk the service runs lexical-only.

Agents use it through a CLI, MCP or HTTP

The Rust/axum server exposes an HTTP API; the kyb CLI, an agent skill and an MCP server sit on top, and a web UI shows the base and its history.

Example uses

01. Service handover

Record verified service changes so the next person or agent can continue the work with current context.

02. Postmortem history

Inspect historical configuration while investigating an incident or postmortem.

03. Incident resolution

Close an incident with a searchable resolution for later operational work.

04. Decisions before asking

Since v0.3.4, kyb confident is a decision journal agents keep while they work: before asking the owner, an agent checks what was decided before about the same thing, and when it asks, it records the question, the answer and later the outcome.

05. A brief before the work

Since v0.3.4, kyb brief lists what is live in the base as pointers: deadlines, open incidents, active work and recently changed knowledge, capped to fit in about 40 lines.

Where it fits

KYB keeps operational notes and their history in Git. Its search index can be rebuilt from that history; optional local embeddings complement lexical search. It stores knowledge and incident resolutions, rather than executing code like prod-code.

Get started

The v0.3.4 installer picks how to run the server (Homebrew or the release archive on Apple Silicon, the Docker image on Linux amd64/arm64), then installs the kyb CLI and the agent skill; KYB_DRY_RUN=1 prints the plan and changes nothing, and KYB_MODE=client installs only the CLI and skill. Semantic search uses an optional local model; BM25 works without it.

One-line installer (since v0.3.4; KYB_DRY_RUN=1 prints the plan first)

curl -fsSL https://kybmemory.com/install.sh | bash

Install the server from crates.io

cargo install knowyourbusiness --locked

Run it (reindexes from Git, listens on 127.0.0.1:9310)

kyb-server

Clone the repository and start the Docker Compose service (data in ./data)

git clone https://github.com/alex09x/kyb && cd kyb
docker compose up -d

README (v0.3.4) →

CLI reference →

HTTP API →

Limitations

  • Usefulness depends on humans or agents recording verified changes.
  • An optional local model is needed for semantic search.
  • Notes are not live monitoring.
  • Release binaries cover Apple Silicon macOS only; Linux, or an Intel Mac with Docker, runs the Docker image (linux/amd64, linux/arm64). There is no Intel macOS binary.
  • A kyb confident precedent is evidence for the agent to judge, not permission; records are matched with the server’s embedding model, falling back to word overlap without one.
Cite this software · archived version v0.3.4

Cite this software

Alexander Panasenko (2026). KYB (Know Your Business): Git-backed knowledge base and incident tracker for AI agents (v0.3.4). Zenodo. https://doi.org/10.5281/zenodo.24095013

This record reflects the archived software version deposited on Zenodo for citation and reproducibility, which may differ from subsequent development releases.