siglume-agent-core 0.2.5

Open-core orchestrator logic for the Siglume API Store agent runtime — tool-manual quality scoring + LLM provider tool adapters + installed-tool prefilter.

siglume-agent-core

Open-core orchestrator logic for the Siglume API Store agent runtime.

This is the public, AGPL-licensed core of the algorithms the Siglume marketplace uses to:

• Score the quality of a publisher's tool manual (tool_manual_validator)
• Build LLM provider tool definitions in Anthropic / OpenAI tool-use format (provider_adapters)

It is the same code that runs in production — extracted from the private monorepo so publishers, contributors, and self-hosters can read, audit, and improve it.

Status: Phase 1 of a staged extraction. Currently exposes Tier A modules (manual quality scoring + provider adapters). Selection scoring (installed_tool_resolver), simulator (dev_simulator), and analytics derivations (seller_analytics) follow in subsequent releases. See ARCHITECTURE.md for the roadmap.

---

Why this exists

The Siglume marketplace agent had been a black box from the publisher side. When a publisher asked "why didn't my API get picked?" or "why is my manual graded B?", the only way to answer was through platform-side reports.

This repository is the direct answer: read the source, run the same scorer locally, contribute improvements as PRs.

The platform itself remains a hosted service (publishers, buyers, payments, identity, deployment infrastructure all stay private). Only the decision logic — how the agent picks tools, how manuals are scored, how provider tool calls are formatted — is open.

Install

pip install siglume-agent-core           # core only
pip install siglume-agent-core[anthropic] # + Anthropic adapter
pip install siglume-agent-core[openai]    # + OpenAI adapter
pip install siglume-agent-core[dev]       # + test/lint deps

What's in this release (v0.2, Tier A + Tier B Phase 1)

siglume_agent_core.tool_manual_validator

The exact same validator Siglume runs to grade publisher-submitted tool manuals (A / B / C / D / F). Use it locally to predict your manual's grade before submission:

from siglume_agent_core.tool_manual_validator import validate_tool_manual, score_manual_quality

manual = {...}  # your tool manual dict

result = validate_tool_manual(manual)
if not result.ok:
    for err in result.errors:
        print(err.code, err.message, err.field)

quality = score_manual_quality(manual)
print(f"Grade {quality.grade} ({quality.overall_score}/100)")
# Platform accepts grade A and B at publish time; C/D/F are rejected.
if quality.grade in ("A", "B"):
    print("Likely publishable — submit when ready.")
else:
    print("Improve before submitting:")
    for s in quality.improvement_suggestions[:3]:
        print(f"  - {s}")

This is byte-equivalent to the server-side scorer. The Siglume monorepo's runtime depends on this PyPI package, so the same code path runs in production. You can verify the claim yourself:

pip install siglume-agent-core
git clone https://github.com/taihei-05/siglume-agent-core
cd siglume-agent-core
pytest tests/test_quality_score_parity.py

The parity test pins score_manual_quality output for four representative manuals against a frozen snapshot in tests/fixtures/expected_scores.json. If your local grade is B, the server grade is B.

siglume_agent_core.installed_tool_prefilter

TF-IDF + cosine similarity scorer that picks the top-N most-relevant tools when an agent has many bound, so the chat system prompt stays within the input token budget. Pure-Python, no embedding service. Same code the platform runs in production:

from siglume_agent_core.installed_tool_prefilter import select_top_tools_for_prompt
from siglume_agent_core.types import ResolvedToolDefinition

# tools is whatever your code resolved from a binding registry.
top = select_top_tools_for_prompt(tools, user_message="translate this to japanese", max_tools=50)
# `top` is a subset of `tools`, ranked by JTBD relevance, original order preserved.

siglume_agent_core.provider_adapters

Provider-specific adapters that convert an internal tool definition + message thread into the format Anthropic's or OpenAI's tool-use API expects, and parse the response back into a uniform shape.

The provider SDKs are optional extras — install only the ones you use:

pip install siglume-agent-core[anthropic]   # + Anthropic SDK
pip install siglume-agent-core[openai]      # + OpenAI SDK

Without the matching extra, importing the adapter raises a clear ImportError telling you which extra to install. Then:

from siglume_agent_core.provider_adapters.anthropic_tools import AnthropicToolAdapter
from siglume_agent_core.provider_adapters.types import ToolMessage

adapter = AnthropicToolAdapter()
turn = adapter.run_turn(
    model="claude-haiku-4-5-20251001",
    messages=[ToolMessage(role="user", content="...")],
    tools=[...],
    max_output_tokens=2048,
    tool_choice="auto",  # "auto" | "any" | "none"
)
print(turn.tool_calls)  # what the LLM picked

tool_choice="none" means no tool use this turn — the adapter elides the tools array entirely, matching the contract you'd expect from OpenAI. Use the same adapter the platform uses, so you can prototype tool-use applications against either provider with consistent behavior.

What's not in this repo

The following stays in the private platform monorepo because exposing them creates security or business risk:

• Authentication / OAuth credential leasing (connected_account_broker)
• Payment processing & wallet signing
• Production database schema & data
• Per-buyer KYC / AML decisioning
• Marketplace pricing & fee logic
• The execution gateway (capability_gateway) — security/policy boundary

See ARCHITECTURE.md for what's planned to come next vs. what stays private.

License

AGPL-3.0-only.

If you self-host the orchestrator, the AGPL terms apply: changes you make to this code that you operate as a network service must be made available under AGPL-3.0 to your users. Commercial licensing for proprietary deployment is available — contact [e-mail].

Contributing

We accept PRs. See CONTRIBUTING.md. The most useful contribution paths today:

• Improve tool_manual_validator heuristics — many graders are currently keyword-rule based; ML-driven or more nuanced scoring is welcome
• Add edge-case tests — anything you've seen the platform mishandle
• Add new provider adapters — Gemini, Mistral, local models

Tracking issue for the broader publisher-dev-tools initiative: siglume-api-sdk#195.