multichat-api (Python)

Helpers for the MultiChat developer API (docs/api/developer.openapi.yaml, v1.0.0). They never sit between you and your provider: convert, optimise, compact or store context here, then send the request with your own key and the provider's own SDK.

from multichat_api import Client
from openai import OpenAI

mc = Client()   # MULTICHAT_API_KEY; MULTICHAT_URL for your own instance
req = mc.optimise(request, format="openai.chat.completions",
                  steps=["trim", "tools", "reasoning", "model"],
                  settings={"turn": 12, "n": 10, "profile": 1})["request"]
reply = OpenAI().chat.completions.create(**req)

Every operation in the spec has a method: convert, sync_encode, sync_decode, optimise, truncate_tool_results, describe_media, compact, start_compact_job, get_job, cancel_job, summarise, contexts (create_context, store_turns, history_tool, retrieve, ...), RAG (create_collection, add_documents, query), embed, count_tokens, usage. Errors raise MultichatError with status and code.

Tests run against the engine itself (MC_ENGINE, default /home/user/mc-engine): pytest.

Licence

Apache License 2.0: see LICENSE, with the attribution notices in NOTICE. Copyright 2026 Digital Deep Ltd.

Licence headers: source files in this package carry no per-file licence header; the LICENSE and NOTICE files at the package root cover every file in it, and are shipped with every published build. New files may start with the two-line header below, and must not carry any other licence. If you copy files out of this package, keep LICENSE and NOTICE with them (Apache-2.0, section 4). Contributions are accepted under Apache-2.0 (section 5).

# Copyright 2026 Digital Deep Ltd
# SPDX-License-Identifier: Apache-2.0