.png)
4 months ago
4
Your AI non-agent.
A wrapper around siesta scripts to automate workflows.
Be carefull to not push company data in here.
| pick-ticket | Implement a ticket in Jira |
| pick-comment | Change local files to comply with request in GitHub comment |
| ... | ...add more here... |
git clone [email protected]:Spenoki-Solutions-GmbH/slopify.git
export PATH="/path/to/slopify/bin:$PATH"
- Setup: jira-cli
- Setup: GitHub CLI
- Setup: Ollama
- Install siesta: pip3 install git+https://github.com/ihucos/siesta.git
- Run: brew install fzf jq
# ~/.slopify.ini
[jira]
org=jira_org_name
[jira_labels]
label1=owner/repo1
label2=owner/repo2
Just hit me on chat and I send you the file.
- No literally random arbitrary shell command execution (A traditional no-go in software development)
- Code can be audited
- No sandboxing needed
- It's just a few hundert lines of script.
- Forseable costs. A fixed number of API calls to the big models.
- Certain LLM completions can be handled locally.
- Hand-pick the most economical model for the job.
- Error cases can be handled sanely at all.
- Precisely curate the context to the LLMs -> Better performance.
- Plan beforehand the best way to accomplish a task.
- No Vendor lock-in - plug and play different models depending on your needs.
- Discoverability: users can see what your tool can do.
- No need for users to oversee and rubber-stamp AIs intermediate steps.
- Apply the "Rule of three". Copy and paste three times before refactoring.
- Use the smallest model fit for the job.
- Use local models when practicable.
- Balance development speed, completion quality and model size.
- Use clear common sense prompts that solve one task at a time.
- Proper prompt formating does matter.
- This is a general but opiniated solution.
