seed-protocol-core

Open schema validator + CLI for the SEED Protocol AI personality format.

A SEED file is a single portable JSON/YAML document that defines an AI's complete identity, voice, bond, capabilities, operations, and (optionally) 60+ philosophical "soul" sections grounded in Indian darshana.

This package is the public, AGPL-3.0 layer — the schema, the validator, the diff engine, the prompt exporter, and a basic CLI. Premium engines (the actual "AI brain") live in the proprietary SaaS at seed.abhibots.com.

Install

pip install seed-protocol-core

Quickstart

# Create a basic SEED interactively
seed create

# Validate against the v3 JSON Schema
seed validate examples/minimal.seed.json

# Compare two SEED files
seed diff examples/sage.seed.json examples/warrior.seed.json

# Export to a system prompt for any LLM provider
seed export examples/sage.seed.json --format openai
seed export examples/sage.seed.json --format anthropic
seed export examples/sage.seed.json --format google

# Inspect a SEED file
seed info examples/minimal.seed.json

Python API

from seed_protocol_core import SeedFile, SeedDiff, export_prompt

seed = SeedFile.load("examples/minimal.seed.json")

# Validate
errors = seed.validate()           # list[str], empty = valid
print(seed.is_valid())             # True / False

# Render to a system prompt
system_text = export_prompt(seed, format="anthropic")

# Diff two seeds
a = SeedFile.load("a.seed.json")
b = SeedFile.load("b.seed.json")
d = SeedDiff(a.data, b.data)
print(d.format_text())

What's in core

What's NOT in core (hosted engines — seed.abhibots.com)

The premium CLI commands (seed apply, seed evolve, seed darpan, seed load) print a friendly message pointing to the SaaS:

$ seed apply my-soul.seed.json
x  `seed apply` runs on the hosted SEED engines, not in this open core.

   It needs the Dharana selective-loading engine.
   Create your soul (free) at https://seed.abhibots.com/login.
   Or use the MCP wrapper: npx -y seed-protocol-mcp@latest

Why AGPL-3.0?

We picked AGPL because it protects the open spec from being silently proprietary-forked into a closed SaaS clone. If you run a modified copy as a hosted service, AGPL requires you to release your modifications. That keeps the schema open and the playing field fair. If you want to build a proprietary product on top of the engines, license commercially through seed.abhibots.com.

The MCP wrapper (seed-protocol-mcp) remains MIT — that one's just a thin client.

Compatibility — moving from seed-protocol

If you used the previous seed-protocol v2.x PyPI package, the public API surface is preserved — SeedFile, validate_seed, SeedDiff, export_prompt. Engine imports (Dharana, Kavach, Atman, etc.) now live inside the SaaS only. Use the SaaS API or the MCP wrapper to access those.

# Before:
from seed_protocol import SeedFile, DharanaEngine   # DharanaEngine is gone

# After:
from seed_protocol_core import SeedFile             # works
# Dharana — call the SaaS or MCP wrapper instead.

Trademarks

"SEED Protocol", "Avni", "Shilpi", "Vidya", and the Pantheon name combinations are trademarks of Abhiraj Singh. AGPL covers the source code only — see NOTICE for details. Forks must rename.

Contact

• Site: https://seed.abhibots.com
• Start: https://seed.abhibots.com/login (free)
• Connect: https://seed.abhibots.com/connect
• MCP: npx -y seed-protocol-mcp@latest
• Email: abhi@seed.abhibots.com