loremem 0.1.0

Lore SDK — inject organizational memory into any AI agent

loremem — Python SDK for Lore

Inject your company's organizational memory into any AI agent in 3 lines of code.

Lore captures every human correction of an AI output, structures it into a company knowledge graph, and feeds it back to your AI agents so they stop making the same mistakes twice.

loremem is the Python SDK for accessing Lore's Context Injection API.

---

Installation

# From PyPI (when published)
pip install loremem

# From GitHub (MVP — no PyPI publish required)
pip install git+https://github.com/mr-shakib/lore#subdirectory=sdk/python

---

Quickstart (3 minutes)

from loremem import LoreClient

client = LoreClient(
    api_key="sk-lore-xxxx",          # from POST /v1/auth/api-keys
    workspace_id="ws_yourworkspace",  # your Lore workspace ID
)

# ── Step 1: Get context before your LLM call ──────────────────────────────────

ctx = client.get_context(
    query="Draft an MSA for Acme Corp",
    tool="contract-drafting-agent",
    hints={"jurisdiction": "US", "customer_tier": "enterprise"},
    entities=["Acme Corp"],
)

# Prepend to your system prompt
system_prompt = ctx.formatted_injection + "\n\n" + YOUR_BASE_SYSTEM_PROMPT
response = openai_client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "system", "content": system_prompt}, ...],
)

# ── Step 2: Report corrections so Lore learns ─────────────────────────────────

# When a human edits the AI output:
client.report_correction(
    ai_output_id="draft_acme_msa_v1",
    summary="Changed indemnity clause from UK to US_STANDARD template",
    tool="contract-drafting-agent",
    context_tags={"customer": "Acme Corp", "document_type": "MSA"},
    actor_id="[e-mail]",
)

# When a human approves the AI output without changes (positive signal):
client.report_output(
    output_id="draft_acme_msa_v2",
    tool="contract-drafting-agent",
    summary="MSA draft approved — no changes needed",
    actor_id="[e-mail]",
)

After a few corrections, Lore automatically proposes rules like:

"US clients require the US_STANDARD indemnity template"

Confirm the rule once → it's injected into every future AI call automatically.

---

Async usage

For async agent frameworks (LangChain, CrewAI, FastAPI-based agents):

from loremem import AsyncLoreClient

client = AsyncLoreClient(api_key="sk-lore-xxxx", workspace_id="ws_acme")

ctx = await client.get_context(
    query="Route this support ticket",
    tool="support-triage-agent",
)

await client.report_correction(
    ai_output_id="ticket_001",
    summary="Re-routed from Tier 1 to Enterprise team",
    tool="support-triage-agent",
)

---

Never-throw guarantee

Every method in LoreClient and AsyncLoreClient is designed to never raise exceptions. If Lore is unavailable, misconfigured, or rate-limited:

• get_context() returns an empty ContextResponse (.formatted_injection == "")
• report_correction() and report_output() return ReportResult(accepted=False)
• A WARNING is logged via Python's standard logging module

Lore's unavailability will never cause your AI agent to break.

import logging
logging.getLogger("loremem").setLevel(logging.WARNING)  # optional: see SDK warnings

---

API reference

LoreClient(api_key, workspace_id, base_url?)

Parameter	Type	Description
api_key	str	Lore API key (sk-lore-...)
workspace_id	str	Your workspace ID
base_url	str	Default: production Lore API. Set to http://localhost:8000 for local dev

get_context(query, tool, hints?, entities?, max_rules?, max_tokens?)

Returns a ContextResponse:

Field	Type	Description
formatted_injection	str	Ready-to-use string — prepend to system prompt
context_id	str	Unique ID for this context response
rules	list[dict]	Active rules that matched
entities	list[dict]	Entity profiles that matched
decisions	list[dict]	Decision records that matched
cached	bool	True if served from 15-min cache

report_correction(ai_output_id, summary, tool, context_tags?, actor_id?)

Call when a human edits or overrides an AI output.

report_output(output_id, tool, summary?, context_tags?, actor_id?)

Call when a human approves an AI output unchanged (positive signal).

---

Getting an API key

# Create a key (requires Clerk JWT from the dashboard, or bootstrap via Supabase directly)
curl -X POST https://lore-m0st.onrender.com/v1/auth/api-keys \
  -H "Authorization: Bearer <clerk_jwt>" \
  -H "Content-Type: application/json" \
  -d '{"name": "Production SDK key"}'

---

Local development

client = LoreClient(
    api_key="sk-lore-xxxx",
    workspace_id="ws_test",
    base_url="http://localhost:8000",  # local FastAPI server
)

---

License

MIT