loregpt 0.1.2

Python SDK for Lore — coordination memory for multi-agent AI systems: create runs, write events, and fetch read-your-writes context packs.

loregpt

Python SDK for Lore — the open-source coordination memory layer for multi-agent AI systems. Create runs, write events, and fetch read-your-writes context packs.

🚧 Pre-release. Lore is building in the open toward v0.1; this SDK tracks the live API.

Compatibility. SDK 0.1.x targets the Lore v1 API (server ≥ 0.0.1). The SDK and the server version independently — match on the API version, not the release number.

Install

pip install loregpt

Requires Python 3.10+. One dependency, httpx, for connection pooling and a shared sync/async transport — the de-facto standard of the AI-Python ecosystem.

Usage

lore = LoreClient(api_key=api_key)

run = lore.create_run()
result = lore.write(
    run_id=run.run_id,
    agent_id="researcher",
    content="Auth flow moved to v2 - PR #42 merged",
)

pack = lore.pack(
    run_id=run.run_id,
    query="current state of auth work",
    scopes={"team": "platform"},
    min_seq=result.seq,
    token_budget=2000,
)

covered_seq = pack.covered_seq  # >= result.seq once distilled; until then the write is in the raw tail
saved_tokens = pack.saved_tokens  # a coarse estimate vs raw history; small packs may round to 0

write takes exactly one of content (a string, wrapped as {"content": ...}) or payload (an opaque dict); write_state(run_id=..., agent_id=..., entity=..., predicate=..., value=...) writes one working-memory fact seen immediately by a same-run reader.

Async

AsyncLoreClient has the same methods with await:

async with AsyncLoreClient(api_key=api_key) as lore:
    run = await lore.create_run()
    result = await lore.write(run_id=run.run_id, agent_id="researcher", content="hello")
    pack = await lore.pack(run_id=run.run_id, query="state of work", min_seq=result.seq)

Errors

Every failure raises a LoreError. Server errors carry the machine code (discriminate on the class or .code, never string-match) and .http_status; transport and parse failures are distinct classes. Requests are not retried (a write is not idempotent) — a failure surfaces so the caller decides.

from loregpt import LoreError, NotFoundError

try:
    pack = lore.pack(run_id=run_id, query="...", min_seq=999)
except NotFoundError:
    ...  # unknown run, or not this key's project
except LoreError as err:
    print(err.code)  # e.g. "min_seq_out_of_range", "model_mismatch", "unauthorized", "connection"

LoreClient(api_key, *, base_url="http://localhost:8080", timeout=..., headers=..., transport=...) — pass a custom httpx transport for a proxy, instrumentation, or tests.

• Website: https://loregpt.ai
• Source & RFCs: https://github.com/lore-gpt/lore

License: Apache-2.0