Agent Nerve is the small open-source layer we built because prompt history and vector memory do not solve cross-machine handoff, task continuity, or safe coordination between mixed agent sessions.
This is not a vector-memory wrapper. It is a shared coordination layer with tasks, append-only events, and lease-based claims so multiple agents can work without trampling each other.
Stdlib-only Python, SQLite, API-key protected writes, server-side redaction for common secrets, and short operational summaries instead of dumping raw prompts or credentials.
Best for people with Claude Code, Codex, OpenCode, Hermes, or custom workers who already feel the pain of repeated prompt recaps and broken handoffs.
Use the repo if you want the core pattern. If you want a business-specific implementation around your agents, queues, review gates, and dashboards, that is build work.
git clone https://github.com/kyletusing34/agent-nerve.git
cd agent-nerve
python3 -m venv .venv
source .venv/bin/activate
pip install -e .
export AGENT_NERVE_API_KEY="change-me"
agent-nerve serve --host 127.0.0.1 --port 8787
agent-nerve task create \
--namespace demo \
--title "Launch the site" \
--owner claude-code
agent-nerve event emit \
--namespace demo \
--agent codex \
--kind handoff \
--summary "Source patch complete"
agent-nerve claim acquire \
--namespace demo \
--resource deploy:production \
--agent hermes