Skip to content

agent_types.md

Latest commit

 

History

History
68 lines (44 loc) · 2 KB

agent_types.md

File metadata and controls

68 lines (44 loc) · 2 KB
Raw

Agent types

pipal supports agent types, which let you scaffold agents with different default behaviors, templates, and configuration.

Available types

  • default — the standard pipal agent template (general‑purpose assistant).
  • kbchat — a knowledge‑base agent that is wired to a local KB path for grounded Q&A.

You can add more types by creating new templates in the templates directory (see “How types work”).

Create an agent with a specific type

pipal agent create <name> --type <type>

Examples:

# Default type (explicit)
pipal agent create momo --type default

# KB chat type
pipal agent create erwin --type kbchat --kb /tmp/knowledge_base

How types work (behind the scenes)

Agent types are implemented as template folders. Each type defines a set of “core files” that are copied into the agent’s home directory at creation time.

  • Templates live under:
    • src/pipal/templates/<type>/
  • Core files include:
    • AGENTS.md, IDENTITY.md, POLICY.md, USER.md, MEMORY.md
    • type‑specific files (e.g. KB.md for kbchat)

When you run pipal agent create, pipal selects the template folder for the given --type and copies those files into:

~/.pipal/agents/<agent_name>/

For kbchat, the --kb flag is used to populate the KB path, which is written into:

~/.pipal/agents/<agent_name>/KB.md

KB Chat

Create a knowledge‑base agent (kbchat):

pipal agent create erwin --type kbchat --kb /tmp/knowledge_base

This writes the KB path into:

~/.pipal/agents/erwin/KB.md

Model requirements

kbchat agents rely on tool calling (read, grep, find, ls) to navigate the knowledge base. This requires a model with strong tool-calling support. Small models (< 7B parameters) may ignore tool instructions and hallucinate answers instead of reading the KB.

Recommended: use a capable model (e.g. Claude, GPT, or Ollama models ≥ 20B with tool-calling support like qwen3-coder:30b or gpt-oss:20b).