agent-chassis

A sketch of a chassis for human-in-the-loop agents: graph runs, tool policy, approval gates, a knowledge ledger, and SQLite snapshots.

This is a design experiment. The code is a small set of Python primitives, not a product you should build on.

What it sketches

Piece	Idea
models	Agents, runs, graphs, tools, events, artifacts, gates, knowledge, jobs
graph_registry	Compile a declared graph (nodes / edges / routes) to LangGraph
tool_registry	Register tools, risk (read / write / publish), enable/disable, approval
human_gate	Standard review: approve, reject, edit, score, need more data
knowledge_ledger	Candidate → stable / rejected (promote needs confidence ≥ 0.78 and ≥ 2 evidence)
scheduler	Manual / cron / event. Cron is *, */N, or an exact field
store	SQLite WAL: full run JSON plus query tables

The application still owns UI, domain models, and real storage. Node and route handlers stay in app code.

Quick start

git clone https://github.com/bunnysi/agent-chassis.git
cd agent-chassis
python3 -m venv .venv
source .venv/bin/activate
pip install -e '.[test]'
pytest -q

Python ≥ 3.11. Depends on pydantic>=2 and langgraph>=0.2.

Usage

from agent_chassis import ToolRegistry
from agent_chassis.models import ToolDefinition, ToolRisk, ToolExecutionRequest

registry = ToolRegistry(
    tools=[ToolDefinition(id="echo", name="Echo", description="Echo input", risk=ToolRisk.read)],
    handlers={"echo": lambda payload: {"echo": payload}},
)
execution = registry.create_execution(
    ToolExecutionRequest(tool_id="echo", input={"hello": "world"})
)

A graph compile + invoke lives in examples/.

SQLite default path is data/agent-chassis.sqlite3. Override with AGENT_CHASSIS_DB_PATH.

License

MIT