Buffaly user guide
Buffaly isn't here to plan your niece's quincea�era. There are plenty of consumer AI agents for that. Buffaly is an enterprise-grade platform built for high-talent individuals and engineering teams to build upon.
The Buffaly Philosophy
Intelligence that compounds
Standard agents start from scratch every session, forcing you to rely on clunky CLI tools or generic MCP integrations. Buffaly is different. You teach it about your code, your APIs, your tools, and your workflows. Ask it to remember actions. Let it write and add new code directly into its runtime. You don't reorganize your infrastructure around the AI�Buffaly evolves around you. It isn't perfect out of the box, but with your guidance, it self-programs and gets better the more you use it.
No training wheels
There are very few safety rails built into Buffaly. This is not for casual users. Buffaly assumes you know what you want to do and will execute the actions you authorize. Best practice: Only operate Buffaly in environments that are connected to Git or securely backed up. You're an adult, and Buffaly treats you like one.
New to Buffaly
Build the mental model first, then try a few safe workflows.
Operating Buffaly
Learn safety, session workflow, logs, validation, and recovery paths.
Extending Buffaly
Choose the right boundary before adding tools, services, providers, or ontology entries.
Topic index
Tools, services, memory, and skills
Providers, UI, and ontology
Daily use: A better way to work
Using Buffaly is different from chatting with a standard LLM. Treat it like an operational teammate: give it goals, ask for evidence, and use its state-management tools.
Declare intent, not scripts
Give Buffaly the exact outcome you want. Let its semantic engine figure out the intermediate steps and tools unless a strict sequence is required.
Trust the discovery engine
Don't force-feed API endpoints or tool names in your prompts. Let Buffaly natively discover actions and entities from its ontology.
Manage state actively
For long-running tasks, use Plan, Scratch, and Task Artifacts to keep the session organized and prevent context bloat.
Demand evidence
Never assume it worked. Always review Buffaly's validation steps, path checks, and commit logs before approving operational changes.
When you are ready to extend Buffaly
Don't write C# if you don't have to. Choose the smallest, most durable capability boundary that fits your workload before adding code or expanding the ontology. Follow this escalation path:
Semantic Memory No Code
Use structured memory to teach Buffaly durable facts, preferences, contacts, and reusable context.
Prompt Skills Low Code
Group repeatable, text-guided workflows into a Skill so you don't have to rewrite the same instructions every day.
ProtoScript Actions Typed Logic
Write formal, typed actions in ProtoScript when a workflow needs strict operational boundaries, data shaping, or native tool exposure.
Services Native Code
Build C# services when your capabilities require strict identity tracking, lifecycle management, statefulness, or heavy integration with external systems.