JSON

Copy

1{
2    "name": "AgentNative Data Exchange",
3    "description": "Normalized official government and public datasets across federal, state and city agencies, census, police, education and schools, with provenance, aggregations, insights and free samples.",
4    "url": "https://agentnative.cazimedia.com/a2a",
5    "version": "0.3.0",
6    "supportedInterfaces": [
7        {
8            "url": "https://agentnative.cazimedia.com/a2a",
9            "protocolBinding": "JSONRPC",
10            "protocolVersion": "1.0"
11        },
12        {
13            "url": "https://agentnative.cazimedia.com",
14            "protocolBinding": "HTTP+JSON",
15            "protocolVersion": "1.0"
16        }
17    ],
18    "capabilities": {
19        "streaming": false,
20        "pushNotifications": false,
21        "extendedAgentCard": false
22    },
23    "defaultInputModes": [
24        "text/plain",
25        "application/json"
26    ],
27    "defaultOutputModes": [
28        "text/plain",
29        "application/json"
30    ],
31    "skills": [
32        {
33            "id": "public-data-search",
34            "name": "Search official government and public datasets",
35            "description": "Discover federal, state, city, census, police, education and school data with official provenance.",
36            "tags": [
37                "government data",
38                "public datasets",
39                "statistics",
40                "provenance"
41            ],
42            "examples": [
43                "Find official datasets about public schools."
44            ]
45        },
46        {
47            "id": "on-demand-materialization",
48            "name": "Make discovered data query-ready",
49            "description": "Prioritize a compatible official dataset for detached ingestion and poll the durable task through completion.",
50            "tags": [
51                "materialization",
52                "async task",
53                "data import"
54            ],
55            "examples": [
56                "Materialize disc_0123456789abcdef01234567 and tell me when it is query-ready."
57            ]
58        },
59        {
60            "id": "imported-data-sample",
61            "name": "Sample normalized government datasets free",
62            "description": "Inspect official rows, summaries, deterministic insights, freshness and provenance before payment.",
63            "tags": [
64                "free sample",
65                "open data",
66                "normalized records"
67            ],
68            "examples": [
69                "Show me useful free data with provenance."
70            ]
71        },
72        {
73            "id": "imported-data-query",
74            "name": "Query and aggregate official datasets",
75            "description": "Filter, group and aggregate normalized public records across government agencies.",
76            "tags": [
77                "aggregation",
78                "analysis",
79                "official records"
80            ],
81            "examples": [
82                "Group these public records by agency."
83            ]
84        },
85        {
86            "id": "coverage-status",
87            "name": "Inspect source coverage",
88            "description": "Inspect current catalog traversal, materialization, queue, failure, and publication freshness state.",
89            "tags": [
90                "coverage",
91                "freshness",
92                "data sources"
93            ],
94            "examples": [
95                "Which official sources are currently queryable?"
96            ]
97        },
98        {
99            "id": "federal-register-query",
100            "name": "Query Federal Register",
101            "description": "Query normalized rules, proposed rules, notices, and presidential documents with bounded date and type filters.",
102            "tags": [
103                "Federal Register",
104                "regulations",
105                "rules",
106                "notices"
107            ],
108            "examples": [
109                "Give me a free current Federal Register briefing with provenance."
110            ]
111        }
112    ]
113}

PYTHON

Copy

1client = await create_client("https://agentnative.cazimedia.com", ClientConfig(streaming=False))
2message = Message(message_id=str(uuid.uuid4()), role=Role.ROLE_USER,
3                  parts=[Part(text="Give me a free current Federal Register briefing with provenance.")])
4async for event in client.send_message(SendMessageRequest(message=message)):
5    ...

BASH

Copy

1curl -X POST https://agentnative.cazimedia.com/a2a \
2  -H "Content-Type: application/json" -H "A2A-Version: 1.0" \
3  -d '{"jsonrpc":"2.0","id":1,"method":"SendMessage","params":{"message":{"role":"ROLE_USER","message_id":"wire-check-1","parts":[{"text":"Give me a free current Federal Register briefing with provenance."}]}}}'

JSON

Copy

1{"jsonrpc": "2.0", "id": 1, "result": {"message": {"messageId": "...", "role": "ROLE_AGENT", "parts": [{"text": "<a JSON document, shown next>"}]}}}

JSON

Copy

1{
2  "type": "federal-register-briefing",
3  "status": 200,
4  "data": {
5    "dataset": "us-federal-register-documents",
6    "sample": true,
7    "sample_basis": "Aggregates cover the 25 newest documents; three example records are included.",
8    "rows": [
9      {
10        "document_number": "2026-19251",
11        "title": "Presidential Determination on Major Drug Transit or Major Illicit Drug Producing Countries ...",
12        "publication_date": "2026-09-18",
13        "type": "Presidential Document",
14        "agencies": [
15          "Executive Office of the President"
16        ],
17        "abstract": null,
18        "url": "https://www.federalregister.gov/documents/2026/09/18/2026-19251/presidential-determi ..."
19      },
20      {
21        "document_number": "2026-19222",
22        "title": "Employment in the Excepted Service",
23        "publication_date": "2026-09-18",
24        "type": "Proposed Rule",
25        "agencies": [
26          "Personnel Management Office"
27        ],
28        "abstract": "The Office of Personnel Management (OPM) proposes to amend its regulations governing the e ...",
29        "url": "https://www.federalregister.gov/documents/2026/09/18/2026-19222/employment-in-the-ex ..."
30      },
31      {
32        "document_number": "2026-19211",
33        "title": "International Traffic in Arms Regulations: Modification of U.S. Munitions List Category XX ...",
34        "publication_date": "2026-09-18",
35        "type": "Rule",
36        "agencies": [
37          "State Department"
38        ],
39        "abstract": "The Department of State (the Department) amends the International Traffic in Arms Regulati ...",
40        "url": "https://www.federalregister.gov/documents/2026/09/18/2026-19211/international-traffi ..."
41      }
42    ],
43    "insights": {
44      "documents_analyzed": 25,
45      "document_mix": {
46        "Presidential Document": 1,
47        "Proposed Rule": 1,
48        "Rule": 2,
49        "Notice": 21
50      },
51      "activity_by_publication_day": {
52        "2026-09-18": 25
53      },
54      "top_agencies": [
55        {
56          "name": "Energy Department",
57          "documents": 7
58        },
59        {
60          "name": "Federal Energy Regulatory Commission",
61          "documents": 7
62        },
63        {
64          "name": "Federal Reserve System",
65          "documents": 2
66        },
67        {
68          "name": "Homeland Security Department",
69          "documents": 2
70        },
71        {
72          "name": "United States Sentencing Commission",
73          "documents": 2
74        }
75      ],
76      "abstract_coverage": "...",
77      "signals": {
78        "most_active_agency": "Energy Department",
79        "dominant_document_type": "Notice"
80      }
81    },
82    "provenance": "https://www.federalregister.gov/developers/documentation/api/v1",
83    "upgrade": {
84      "options": {
85        "pass": "$5 for 250 queries / 30 days",
86        "subscription": "$5/month for 2,000 queries"
87      },
88      "benefits": [
89        "up to 100 rows per query",
90        "date and document-type filters",
91        "aggregations and provenance",
92        "120 requests per minute"
93      ],
94      "access_start": "https://agentnative.cazimedia.com/v1/access/start"
95    },
96    "rate_limit": {
97      "limit": 20,
98      "remaining": 19,
99      "reset": 1789882740
100    }
101  }
102}

PYTHON

Copy

1import asyncio
2import json
3import operator
4import re
5import sys
6import uuid
7from typing import Annotated, TypedDict
8
9import httpx
10from a2a.client import A2ACardResolver, ClientConfig, create_client
11from a2a.types import Message, Part, Role, SendMessageRequest, TaskState
12from langchain_ollama import ChatOllama
13from langgraph.checkpoint.memory import MemorySaver
14from langgraph.graph import END, START, StateGraph
15from langgraph.types import Command, Send, interrupt
16from pydantic import BaseModel
17
18AGENT_URL = "https://agentnative.cazimedia.com"
19SKILL_ID = "federal-register-query"
20BRIEFING_PROMPT = "Give me a free current Federal Register briefing with provenance."
21OFFICIAL_API = "https://www.federalregister.gov/api/v1/documents/{number}.json"
22OFFICIAL_FIELDS = ["document_number", "title", "type", "publication_date", "html_url", "agencies", "abstract"]
23MODEL = "qwen2.5:14b"
24DEFAULT_ROLE = "export compliance officer at a manufacturer of underwater and defense electronics"
25RECIPIENT = "Dana Okafor, general counsel"
26
27BAD_STATES = {TaskState.TASK_STATE_FAILED, TaskState.TASK_STATE_REJECTED,
28              TaskState.TASK_STATE_INPUT_REQUIRED, TaskState.TASK_STATE_AUTH_REQUIRED}
29
30
31class State(TypedDict):
32    role: str
33    card: dict
34    rows: list[dict]
35    insights: dict
36    checked: Annotated[list[dict], operator.add]  # parallel branches append here
37    picks: list[dict]
38    note: str
39    decision: str
40
41
42class Score(BaseModel):
43    document_number: str
44    relevance: int
45    reason: str
46
47
48class Scores(BaseModel):
49    scores: list[Score]
50
51
52def reply_text(event) -> str:
53    kind = event.WhichOneof("payload")
54    if kind == "message":
55        return "".join(p.text for p in event.message.parts)
56    if kind == "task":
57        if event.task.status.state in BAD_STATES:
58            raise RuntimeError(f"task ended {TaskState.Name(event.task.status.state)}")
59        return "".join(p.text for a in event.task.artifacts for p in a.parts)
60    return ""
61
62
63async def discover(state: State) -> dict:
64    async with httpx.AsyncClient() as http:
65        card = await A2ACardResolver(http, AGENT_URL).get_agent_card()
66    skills = [s.id for s in card.skills]
67    if SKILL_ID not in skills:
68        raise RuntimeError(f"{card.name} does not advertise {SKILL_ID}")
69    iface = card.supported_interfaces[0]
70    print(f"  [discover] {card.name} v{card.version}: {iface.protocol_binding} "
71          f"{iface.protocol_version}, {len(skills)} skills, {SKILL_ID} found")
72    return {"card": {"name": card.name, "skills": skills}}
73
74
75async def fetch_briefing(state: State) -> dict:
76    client = await create_client(AGENT_URL, ClientConfig(streaming=False))
77    text = ""
78    try:
79        message = Message(message_id=str(uuid.uuid4()), role=Role.ROLE_USER,
80                          parts=[Part(text=BRIEFING_PROMPT)])
81        async for event in client.send_message(SendMessageRequest(message=message)):
82            text += reply_text(event)
83    finally:
84        await client.close()
85
86    reply = json.loads(text)
87    if reply.get("type") != "federal-register-briefing" or reply.get("status") != 200:
88        raise RuntimeError(f"unexpected reply: {text[:200]}")
89    data = reply["data"]
90
91    # The agent is an unverified third party: keep only the fields this graph needs,
92    # in the shape it expects. The upgrade offers and everything else are dropped.
93    rows = []
94    for r in data["rows"]:
95        if (re.fullmatch(r"\d{4}-\d{4,6}", str(r.get("document_number", "")))
96                and str(r.get("url", "")).startswith("https://www.federalregister.gov/")):
97            rows.append({"document_number": r["document_number"], "title": str(r.get("title", ""))[:300],
98                         "type": str(r.get("type", "")), "publication_date": str(r.get("publication_date", "")),
99                         "url": r["url"]})
100    ins = data.get("insights", {})
101    insights = {"documents_analyzed": ins.get("documents_analyzed"), "document_mix": ins.get("document_mix"),
102                "top_agencies": ins.get("top_agencies", [])[:5]}
103    remaining = data.get("rate_limit", {}).get("remaining")
104    print(f"  [fetch_briefing] {len(rows)} rows kept of {len(data['rows'])}; "
105          f"insights cover {insights['documents_analyzed']} documents; free-tier calls remaining: {remaining}")
106    return {"rows": rows, "insights": insights}
107
108
109def fan_out(state: State) -> list[Send]:
110    return [Send("verify_document", {"row": r}) for r in state["rows"]]
111
112
113def squash(s: str) -> str:
114    return re.sub(r"\s+", " ", s).strip()
115
116
117async def verify_document(payload: dict) -> dict:
118    row = payload["row"]
119    params = [("fields[]", f) for f in OFFICIAL_FIELDS]
120    async with httpx.AsyncClient(timeout=30) as http:
121        resp = await http.get(OFFICIAL_API.format(number=row["document_number"]), params=params)
122    problems, official = [], None
123    if resp.status_code == 404:
124        problems.append("not found at federalregister.gov")
125    else:
126        resp.raise_for_status()
127        official = resp.json()
128        for field, official_field in (("title", "title"), ("type", "type"),
129                                      ("publication_date", "publication_date"), ("url", "html_url")):
130            if squash(row[field]) != squash(official[official_field]):
131                problems.append(f"{field} differs from the official record")
132    ok = not problems
133    print(f"  [verify_document] {row['document_number']}: {'matches the official record' if ok else '; '.join(problems)}")
134    result = {"document_number": row["document_number"], "ok": ok, "problems": problems}
135    if ok:
136        result["official"] = {
137            "title": official["title"], "type": official["type"], "date": official["publication_date"],
138            "agencies": [a["name"] for a in official["agencies"]], "url": official["html_url"],
139            "abstract": (official.get("abstract") or "")[:700]}
140    return {"checked": [result]}
141
142
143def score_relevance(state: State) -> dict:
144    good = {c["document_number"]: c["official"] for c in state["checked"] if c["ok"]}
145    if not good:
146        return {"picks": []}
147    docs = "\n\n".join(
148        f"{n}: {d['title']}\nType: {d['type']}. Agencies: {', '.join(d['agencies'])}. Published {d['date']}.\n"
149        f"Abstract: {d['abstract'] or '(none published)'}" for n, d in good.items())
150    model = ChatOllama(model=MODEL, temperature=0).with_structured_output(Scores)
151    result = model.invoke(
152        f"I am a {state['role']}. Below are new Federal Register documents. For each one, give a "
153        "relevance score from 0 (irrelevant to my work) to 10 (I must read it today) and a one-sentence "
154        "reason that refers to my role. Use each document_number exactly as given.\n\n" + docs)
155    picks = []
156    for s in result.scores:
157        if s.document_number in good and s.relevance >= 6:
158            picks.append({**good[s.document_number], "document_number": s.document_number,
159                          "relevance": min(s.relevance, 10), "reason": s.reason})
160    for s in result.scores:
161        print(f"  [score_relevance] {s.document_number}: {s.relevance}/10 - {s.reason}")
162    return {"picks": sorted(picks, key=lambda p: -p["relevance"])}
163
164
165def route(state: State) -> str:
166    return "draft_note" if state["picks"] else "nothing_relevant"
167
168
169def draft_note(state: State) -> dict:
170    items = "\n".join(f"- {p['title']} ({p['type']}, {p['date']}). Why it matters: {p['reason']}"
171                      for p in state["picks"])
172    body = ChatOllama(model=MODEL, temperature=0).invoke(
173        f"Write the body of a short note (under 120 words) from Neil, a {state['role']}, to {RECIPIENT}, "
174        "flagging the Federal Register documents below. Do not include links, document numbers, a greeting "
175        f"or a sign-off, and do not add facts that are not listed.\n\n{items}").content.strip()
176    # Links come from the verified official records, never from the model.
177    sources = "\n".join(f"- {p['title']} ({p['type']}, {p['date']}): {p['url']}" for p in state["picks"])
178    note = f"Dear {RECIPIENT.split(',')[0].split()[0]},\n\n{body}\n\nSources:\n{sources}\n\nBest regards,\nNeil"
179    print("  [draft_note] drafted")
180    return {"note": note}
181
182
183def human_review(state: State) -> dict:
184    return {"decision": interrupt({"note": state["note"], "question": "Send this note? (approve / reject)"})}
185
186
187def build_graph():
188    g = StateGraph(State)
189    for name, fn in (("discover", discover), ("fetch_briefing", fetch_briefing),
190                     ("verify_document", verify_document), ("score_relevance", score_relevance),
191                     ("draft_note", draft_note), ("human_review", human_review)):
192        g.add_node(name, fn)
193    g.add_edge(START, "discover")
194    g.add_edge("discover", "fetch_briefing")
195    g.add_conditional_edges("fetch_briefing", fan_out, ["verify_document"])
196    g.add_edge("verify_document", "score_relevance")
197    g.add_conditional_edges("score_relevance", route, {"draft_note": "draft_note", "nothing_relevant": END})
198    g.add_edge("draft_note", "human_review")
199    g.add_edge("human_review", END)
200    return g.compile(checkpointer=MemorySaver())
201
202
203async def run_until_pause(graph, payload, config):
204    async for update in graph.astream(payload, config, stream_mode="updates"):
205        if "__interrupt__" in update:
206            return update["__interrupt__"][0].value
207    return None
208
209
210async def tamper_test() -> None:
211    real = {"document_number": "2026-19211", "title": "International Traffic in Arms Regulations: "
212            "Modification of U.S. Munitions List Category XX(a)", "type": "Rule",
213            "publication_date": "2026-09-18", "url": "https://www.federalregister.gov/documents/2026/09/18/"
214            "2026-19211/international-traffic-in-arms-regulations-modification-of-us-munitions-list-category-xxa"}
215    for label, row in (("untouched row", real),
216                       ("title altered", {**real, "title": "Repeal of all export controls"}),
217                       ("invented document", {**real, "document_number": "2026-99999"})):
218        print(f"{label}:")
219        await verify_document({"row": row})
220
221
222async def main(role: str) -> None:
223    graph = build_graph()
224    config = {"configurable": {"thread_id": "watch-1"}}
225    print(f"role: {role}\n--- run ---")
226    pending = await run_until_pause(graph, {"role": role, "checked": []}, config)
227    if pending is None:
228        print("\n--- finished without pausing: no document was relevant, nothing drafted ---")
229        return
230    print("\n--- paused for human review ---")
231    print(pending["note"])
232    answer = input(f"\n{pending['question']} ").strip() or "reject"
233    print("\n--- resumed ---")
234    await run_until_pause(graph, Command(resume=answer), config)
235    final = (await graph.aget_state(config)).values
236    print(f"decision recorded: {final['decision']}")
237
238
239if __name__ == "__main__":
240    if "--tamper-test" in sys.argv:
241        asyncio.run(tamper_test())
242    elif "--graph" in sys.argv:
243        print(build_graph().get_graph().draw_mermaid())
244    else:
245        asyncio.run(main(next((a for a in sys.argv[1:] if not a.startswith("--")), DEFAULT_ROLE)))

BASH

Copy

1brew install uv

BASH

Copy

1  ollama pull qwen2.5:14b
2  curl http://localhost:11434/api/version

CODE

Copy

1  langgraph
2  langchain-ollama
3  httpx
4  a2a-sdk

BASH

Copy

1uv venv --python 3.12 .venv
2source .venv/bin/activate
3uv pip install -r requirements.txt

BASH

Copy

1python regulatory_watch.py --tamper-test

BASH

Copy

1python regulatory_watch.py                                      # the default role
2python regulatory_watch.py "marine biologist restoring coral reefs"
3echo approve | python regulatory_watch.py                       # answers the approval prompt for you
4python regulatory_watch.py --graph                              # prints the compiled graph as Mermaid text

BASH

Copy

1uv run --python 3.12 --with-requirements requirements.txt regulatory_watch.py --tamper-test

BASH

Copy

1uv run --python 3.12 --with-requirements requirements.txt regulatory_watch.py "marine biologist restoring coral reefs"

CODE

Copy

1untouched row:
2  [verify_document] 2026-19211: matches the official record
3title altered:
4  [verify_document] 2026-19211: title differs from the official record
5invented document:
6  [verify_document] 2026-99999: not found at federalregister.gov

CODE

Copy

1role: export compliance officer at a manufacturer of underwater and defense electronics
2--- run ---
3  [discover] AgentNative Data Exchange v0.3.0: JSONRPC 1.0, 6 skills, federal-register-query found
4  [fetch_briefing] 3 rows kept of 3; insights cover 25 documents; free-tier calls remaining: 19
5  [verify_document] 2026-19222: matches the official record
6  [verify_document] 2026-19211: matches the official record
7  [verify_document] 2026-19251: matches the official record
8  [score_relevance] 2026-19251: 0/10 - This document pertains to drug transit and production and is not directly related to underwater or defense electronics.
9  [score_relevance] 2026-19222: 0/10 - This document concerns employment regulations and does not pertain to export compliance for underwater or defense electronics.
10  [score_relevance] 2026-19211: 10/10 - This document directly affects the export regulations for uncrewed underwater vehicles, which is highly relevant to my role in export compliance for defense electronics.
11  [draft_note] drafted
12
13--- paused for human review ---
14Dear Dana,
15
16Please be advised of a recent publication in the Federal Register regarding the International Traffic in Arms Regulations: Modification of U.S. Munitions List Category XX(a). This modification is crucial as it directly impacts the export regulations for uncrewed underwater vehicles, an area highly relevant to our export compliance for defense electronics. I recommend we review the implications of this change on our current export practices and ensure compliance with the updated regulations.
17
18Sources:
19- International Traffic in Arms Regulations: Modification of U.S. Munitions List Category XX(a) (Rule, 2026-09-18): https://www.federalregister.gov/documents/2026/09/18/2026-19211/international-traffic-in-arms-regulations-modification-of-us-munitions-list-category-xxa
20
21Best regards,
22Neil
23
24Send this note? (approve / reject) approve
25--- resumed ---
26decision recorded: approve

CODE

Copy

1role: marine biologist restoring coral reefs
2--- run ---
3  [discover] AgentNative Data Exchange v0.3.0: JSONRPC 1.0, 6 skills, federal-register-query found
4  [fetch_briefing] 3 rows kept of 3; insights cover 25 documents; free-tier calls remaining: 19
5  [verify_document] 2026-19211: matches the official record
6  [verify_document] 2026-19251: matches the official record
7  [verify_document] 2026-19222: matches the official record
8  [score_relevance] 2026-19251: 0/10 - This document is about drug transit and production and has no direct relation to marine biology or coral reef restoration.
9  [score_relevance] 2026-19222: 0/10 - This document pertains to employment regulations and does not impact marine biology or coral reef restoration efforts.
10  [score_relevance] 2026-19211: 2/10 - This document modifies regulations on uncrewed underwater vehicles, which could indirectly affect marine research and conservation efforts, but is not directly relevant to coral reef restoration.
11
12--- finished without pausing: no document was relevant, nothing drafted ---