@infolang/connectors

Source connectors that push external content into InfoLang hosted memory (https://api.infolang.ai). Currently ships a working GitHub full-repo connector and a Notion connector that's built and unit-tested but not yet wired into the CLI.

Not published to npm yet. Install from source until it is.

Install

git clone https://github.com/InfoLang-Inc/infolang-connectors.git
cd infolang-connectors
npm install
npm link   # optional: puts `infolang-connectors` on your PATH

Requires Node 18+.

Usage

infolang-connectors github sync <owner/repo> [--dry-run] [--confirm]

or, without npm link:

node bin/cli.mjs github sync <owner/repo> [--dry-run] [--confirm]

Example:

export INFOLANG_API_KEY=il_live_...
export INFOLANG_WORKSPACE_ID=3f9c1a2e-...

infolang-connectors github sync my-org/my-repo --dry-run
# review the file/chunk/byte summary and destination, then:
infolang-connectors github sync my-org/my-repo --confirm

The first confirmed sync always previews before sending, regardless of --confirm — you need one extra --confirm (or --yes/-y) run to actually push. --dry-run previews on every run, including later syncs.

On each run, the connector clones (first run) or fetches and diffs (later runs) the target repo, applies .gitignore plus a fixed set of always-excluded directories (node_modules, .git, dist, build, ...), skips detected binaries, chunks the rest (heading-aware for Markdown, boundary-aware for code), and pushes the chunks.

Programmatic use

import { syncGithubRepo } from "@infolang/connectors";

const result = await syncGithubRepo({
  ownerRepo: "my-org/my-repo",
  apiKey: process.env.INFOLANG_API_KEY,
  workspaceId: process.env.INFOLANG_WORKSPACE_ID,
  confirm: true,
});

Other exports (lib/index.mjs): pushChunks / pushViaExecute / pushViaRemember (transport layer), chunkFile / chunkMarkdown / chunkCode (chunking), resolveGithubToken / buildCloneUrl (GitHub auth), and the Notion scaffold (syncNotionWorkspace, richTextToMarkdown, blockToMarkdown, pageToMarkdown, buildAuthorizeUrl, exchangeCodeForToken).

Configuration

| Variable / flag | Required | Notes | | --- | --- | --- | | GITHUB_TOKEN / GITHUB_APP_TOKEN / GH_TOKEN / --token | No | GitHub PAT (classic: repo scope; fine-grained: Contents: Read-only) or App installation token. Public repos work with no token. | | INFOLANG_API_KEY / --api-key | Yes | il_live_* or il_test_* InfoLang API key. | | INFOLANG_WORKSPACE_ID / --workspace-id | Yes | Real workspace UUID — see below. | | --api-base | No | Defaults to https://api.infolang.ai. |

Workspace ID

Every sync needs an explicit, real, account-owned InfoLang workspace ID. There's currently no way for a connector to create or discover a workspace on its own: log into the InfoLang console, create (or pick) a workspace, and copy its UUID into INFOLANG_WORKSPACE_ID or --workspace-id.

An arbitrary string here returns 401 Unauthorized — the API only accepts a workspace the calling key is actually scoped to. Omitting the value entirely is also rejected rather than silently falling back to a default namespace, since that would defeat per-source isolation.

Notion connector

Everything under lib/notion/ (OAuth token exchange, block-children fetch with pagination, rich-text/block → Markdown conversion) is implemented and unit-tested against Notion's REST API, but there's no CLI verb for it yet and it makes no live calls. syncNotionWorkspace() returns { mode: "inactive", reason } until NOTION_CLIENT_ID / NOTION_CLIENT_SECRET are set (create an integration at notion.so/my-integrations) and a token from lib/notion/oauth.mjs's OAuth flow is passed as NOTION_TOKEN / --token.

Tests

npm test

Runs node --test test/*.test.mjs. Everything is mocked — no live network or git calls.

License

MIT — see LICENSE.md.