CogniRelay is a self-hosted continuity and collaboration substrate for autonomous agents with bounded, recoverable memory. It uses a local git repository as durable state, exposes a machine-first FastAPI interface, stores content as Markdown and JSON/JSONL, and keeps dependencies minimal.
CogniRelay is not a Git forge or a full orchestration framework. It is infrastructure for memory, retrieval, messaging, coordination, and continuity preservation across context-window resets. It is agent-agnostic: any runtime that can call the HTTP or MCP surfaces can use it.
The default deployment model is one owner-agent per CogniRelay instance. Collaborating peers can receive narrower API tokens and interact through coordination surfaces without access to the owner’s continuity capsules.
For the full product introduction, installation notes, and feature summary, see the README.
Source and GitHub release installs use a virtual environment, pip install -r
requirements.txt, and python -m uvicorn app.main:app --host 127.0.0.1 --port
8080.
PyPI installs use pip install cognirelay and cognirelay serve --host
127.0.0.1 --port 8080. Set COGNIRELAY_REPO_ROOT to a durable writable
runtime-state directory outside site-packages; the default ./data_repo is
only for local/manual development.
PyPI installs also include the last-mile agent assets under the installed
cognirelay/agent_assets package-data directory. Use cognirelay assets path
to print that installed path, cognirelay assets list to inspect the bundled
allowlist, or cognirelay assets copy --to <dir> to write <dir>/agent-assets
for local agent configuration.
Wheel installs do not bundle the full source documentation in this slice. The
/ui/docs page may show degraded or unavailable doc entries unless
COGNIRELAY_DOCS_SOURCE_ROOT points at a source checkout.
Production deployments can use the templates under deploy/ with either a
source checkout or an installed package. Docker is not the default deployment
path.