cachly-brain-setup

GitHub Action to auto-configure AI Brain memory for any repo.

Usage

name: Setup Cachly Brain
on:
  push:
    branches: [main]

jobs:
  setup-brain:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      - uses: cachly-dev/cachly-action@v1
        with:
          instance-id: ${{ secrets.CACHLY_INSTANCE_ID }}
          api-key: ${{ secrets.CACHLY_API_KEY }}
          project-description: 'My awesome project'

      - uses: stefanzweifel/git-auto-commit-action@v5
        with:
          commit_message: 'chore: configure Cachly AI Brain'
          file_pattern: '.github/copilot-instructions.md .mcp.json'

Modes

setup (default)

Generates .github/copilot-instructions.md and .mcp.json, then indexes your source files into the Brain. Run on every push to main to keep the index fresh.

learn

Auto-learns from recent commits and PR metadata. Wire this to pull_request: types: [closed] so your Brain grows on every merge — no manual learn_from_attempts calls needed.

scan

PR risk scan. Calls brain_plan + smart_recall against the PR diff and posts a comment with a risk score (0–100) and a list of predicted failure patterns. Run this on pull_request: types: [opened, synchronize] before CI runs so reviewers see the risk upfront.

hygiene

Weekly Brain sweep. Archives stale lessons, flags provisional knowledge, and removes orphaned entries. Schedule via workflow_dispatch or a weekly cron — keeps Brain quality high over time.

GitLab CI/CD

On GitLab? The same Brain, same modes, same closed-loop self-calibration are available via the GitLab template at templates/cachly.gitlab-ci.yml. Add to your .gitlab-ci.yml:

include:
  - remote: 'https://raw.githubusercontent.com/cachly-dev/cachly-action/main/templates/cachly.gitlab-ci.yml'

variables:
  CACHLY_INSTANCE_ID: "<your-brain-uuid>"

cachly-learn:
  extends: .cachly_learn
  rules:
    - if: '$CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH'

cachly-scan:
  extends: .cachly_scan
  rules:
    - if: '$CI_PIPELINE_SOURCE == "merge_request_event"'

Set CACHLY_API_KEY (masked) in Settings → CI/CD → Variables. Outcomes are reported to the Brain as source gitlab_ci. The Cachly VS Code and JetBrains plugins auto-detect a GitLab origin and scaffold this for you on setup.

Inputs

Input	Required	Default	Description
instance-id	✅	—	Brain instance UUID (from cachly.dev → Dashboard → Brain)
api-key	✅	—	API key cky_live_... (use GitHub Secrets!)
api-url	❌	https://api.cachly.dev	API base URL
embed-provider	❌	openai	Embedding provider
project-description	❌	repo name	Short project description
index-project	❌	true	Index source files into the Brain semantic cache
index-max-files	❌	500	Max files to index per run
mode	❌	setup	setup (generate config + index), learn (auto-learn from recent commits), scan (PR risk scan), or hygiene (weekly Brain sweep)
learn-max-commits	❌	50	In learn mode: how many recent commits to learn from
pr-number	❌	—	PR number (required for scan mode)
pr-title	❌	—	PR title passed to risk scan
pr-body	❌	—	PR body / description passed to risk scan
scan-top-k	❌	10	In scan mode: number of Brain lessons to consider
scan-post-comment	❌	true	In scan mode: whether to post a PR comment with the risk score

Outputs

Output	Description
indexed-files	Number of files indexed into the Brain
risk-score	Risk score 0–100 produced by scan mode (0 = low risk, 100 = very high)
failures-json	JSON array of predicted failure patterns from scan mode

---

Full workflow example

All four modes wired together in one CI file:

name: Cachly Brain

on:
  push:
    branches: [main]
  pull_request:
    types: [opened, synchronize, closed]
  schedule:
    - cron: '0 3 * * 1'   # weekly hygiene — every Monday at 03:00 UTC
  workflow_dispatch:

jobs:
  # ── setup: keep the Brain index fresh on every push to main ──────────
  setup:
    if: github.event_name == 'push'
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: cachly-dev/cachly-action@v1
        with:
          mode: setup
          instance-id: ${{ secrets.CACHLY_INSTANCE_ID }}
          api-key: ${{ secrets.CACHLY_API_KEY }}
          project-description: 'My awesome project'
      - uses: stefanzweifel/git-auto-commit-action@v5
        with:
          commit_message: 'chore: configure Cachly AI Brain'
          file_pattern: '.github/copilot-instructions.md .mcp.json'

  # ── scan: post PR risk comment before CI runs ─────────────────────────
  scan:
    if: github.event_name == 'pull_request' && github.event.action != 'closed'
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: cachly-dev/cachly-action@v1
        with:
          mode: scan
          instance-id: ${{ secrets.CACHLY_INSTANCE_ID }}
          api-key: ${{ secrets.CACHLY_API_KEY }}
          pr-number: ${{ github.event.pull_request.number }}
          pr-title: ${{ github.event.pull_request.title }}
          pr-body: ${{ github.event.pull_request.body }}
          scan-top-k: 10
          scan-post-comment: 'true'

  # ── learn: auto-learn from every merged PR ────────────────────────────
  learn:
    if: github.event_name == 'pull_request' && github.event.pull_request.merged == true
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 50
      - uses: cachly-dev/cachly-action@v1
        with:
          mode: learn
          instance-id: ${{ secrets.CACHLY_INSTANCE_ID }}
          api-key: ${{ secrets.CACHLY_API_KEY }}

  # ── hygiene: weekly Brain sweep ───────────────────────────────────────
  hygiene:
    if: github.event_name == 'schedule' || github.event_name == 'workflow_dispatch'
    runs-on: ubuntu-latest
    steps:
      - uses: cachly-dev/cachly-action@v1
        with:
          mode: hygiene
          instance-id: ${{ secrets.CACHLY_INSTANCE_ID }}
          api-key: ${{ secrets.CACHLY_API_KEY }}

---

🧠 Auto-learn from every merged PR (zero manual work)

Add this workflow and your Brain grows automatically on every PR merge — each meaningful commit becomes a lesson your AI recalls in future sessions:

name: Brain — learn from merged PRs
on:
  pull_request:
    types: [closed]

jobs:
  learn:
    if: github.event.pull_request.merged == true
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 50   # must be >= learn-max-commits for history access
      - uses: cachly-dev/cachly-action@v1
        with:
          mode: learn
          instance-id: ${{ secrets.CACHLY_INSTANCE_ID }}
          api-key: ${{ secrets.CACHLY_API_KEY }}

After this, every merged PR teaches your Brain — no learn_from_attempts calls, no manual steps. Your AI arrives pre-briefed on what the team shipped.

What it generates

• .github/copilot-instructions.md — Auto-recall/learn instructions for all AI assistants
• .mcp.json — MCP server config (if not present)

After setup, every push automatically indexes your latest source changes into the Brain so your AI assistant can find relevant code with semantic_search — no manual re-indexing needed.

---

👥 Team Brain — Shared AI Memory for Your Whole Team

One shared instance. Every developer gets smarter every day.

# GitHub Actions — share the same Brain instance across the whole team
env:
  CACHLY_INSTANCE_ID: ${{ secrets.CACHLY_INSTANCE_ID }}

# Alice's workflow stores a lesson after a successful deploy:
- name: Store deploy lesson
  run: |
    curl -sX POST https://api.cachly.dev/v1/brain/learn \
      -H "Authorization: Bearer ${{ secrets.CACHLY_API_KEY }}" \
      -d '{"topic":"deploy:k8s-timeout","outcome":"success",
           "what_worked":"Increase readinessProbe.failureThreshold to 10","author":"alice"}'

# Bob's next session-start workflow pulls all team lessons automatically:
- uses: cachly-dev/cachly-action@v1
  with:
    instance-id: ${{ secrets.CACHLY_INSTANCE_ID }}
    api-key: ${{ secrets.CACHLY_API_KEY }}
# → "💡 alice solved deploy:k8s-timeout 1d ago: Increase readinessProbe..."

Set up a team org at cachly.dev/teams — Team €99/mo · 10 seats · Business €299/mo · 50 seats.