Files
foodlinkk-command-center/cockpit/app/services/agent_chat.py
T

590 lines
19 KiB
Python

"""War Room + 1:1 agent chat — Hermes/mimo only, soul prompts, A2A rounds."""
from __future__ import annotations
import asyncio
import logging
import re
from typing import Any
from app.config import settings
from app.db import execute, fetch_all, fetch_one
from app.services import agent_souls, hermes_memory, llm_router
from app.services.agent_names import normalize_agent_key
log = logging.getLogger("cockpit.agent_chat")
MENTION_RE = re.compile(r"@([a-zA-Z][a-zA-Z0-9_-]{1,63})")
MAX_ROUNDS = 2
MAX_REPLIES_PER_ROUND = 3
HISTORY_LIMIT = 16
WARROOM_MAX_TOKENS = 512
WARROOM_TIMEOUT = 55.0
def _iso(row: dict[str, Any] | None) -> dict[str, Any] | None:
if not row:
return None
out = dict(row)
for k, v in list(out.items()):
if hasattr(v, "isoformat"):
out[k] = v.isoformat()
if isinstance(out.get("metadata"), str):
try:
import json
out["metadata"] = json.loads(out["metadata"])
except Exception:
out["metadata"] = {}
return out
def ensure_warroom() -> dict[str, Any]:
row = fetch_one("SELECT * FROM warroom_rooms WHERE kind = 'warroom' LIMIT 1")
if row:
return _iso(row) # type: ignore
row = fetch_one(
"""INSERT INTO warroom_rooms (kind, title) VALUES ('warroom', 'War Room')
RETURNING *"""
)
return _iso(row) # type: ignore
def get_or_create_dm(agent_key: str) -> dict[str, Any]:
key = normalize_agent_key(agent_key) or (agent_key or "").strip().lower()
soul = agent_souls.get_soul(key)
if not soul:
raise ValueError(f"Onbekende agent: {agent_key}")
row = fetch_one("SELECT * FROM warroom_rooms WHERE kind = 'dm' AND agent_key = %s", (key,))
if row:
return _iso(row) # type: ignore
title = f"DM · {soul.get('display_name') or key}"
row = fetch_one(
"""INSERT INTO warroom_rooms (kind, agent_key, title)
VALUES ('dm', %s, %s) RETURNING *""",
(key, title),
)
return _iso(row) # type: ignore
def list_rooms() -> list[dict[str, Any]]:
ensure_warroom()
rows = fetch_all(
"""SELECT r.*,
(SELECT COUNT(*) FROM warroom_messages m WHERE m.room_id = r.id) AS message_count,
(SELECT content FROM warroom_messages m WHERE m.room_id = r.id
ORDER BY m.created_at DESC LIMIT 1) AS last_message
FROM warroom_rooms r
ORDER BY CASE WHEN r.kind = 'warroom' THEN 0 ELSE 1 END, r.title"""
)
return [_iso(r) for r in rows] # type: ignore
def get_room(room_id: int) -> dict[str, Any] | None:
return _iso(fetch_one("SELECT * FROM warroom_rooms WHERE id = %s", (room_id,)))
def list_messages(room_id: int, *, limit: int = 100, after_id: int | None = None) -> list[dict[str, Any]]:
if after_id:
rows = fetch_all(
"""SELECT * FROM warroom_messages
WHERE room_id = %s AND id > %s
ORDER BY created_at ASC, id ASC LIMIT %s""",
(room_id, after_id, limit),
)
else:
rows = fetch_all(
"""SELECT * FROM (
SELECT * FROM warroom_messages WHERE room_id = %s
ORDER BY created_at DESC, id DESC LIMIT %s
) t ORDER BY created_at ASC, id ASC""",
(room_id, limit),
)
return [_iso(r) for r in rows] # type: ignore
def _insert_message(
room_id: int,
*,
sender_type: str,
content: str,
agent_key: str | None = None,
username: str | None = None,
sender_avatar_url: str | None = None,
metadata: dict | None = None,
) -> dict[str, Any]:
import json
row = fetch_one(
"""INSERT INTO warroom_messages
(room_id, sender_type, agent_key, username, sender_avatar_url, content, metadata)
VALUES (%s, %s, %s, %s, %s, %s, %s::jsonb)
RETURNING *""",
(
room_id,
sender_type,
agent_key,
username,
sender_avatar_url,
content,
json.dumps(metadata or {}),
),
)
execute("UPDATE warroom_rooms SET updated_at = NOW() WHERE id = %s", (room_id,))
return _iso(row) # type: ignore
def _active_agent_keys() -> set[str]:
return {str(s.get("agent_key")) for s in agent_souls.list_souls() if s.get("is_active", True)}
def parse_mentions(text: str) -> list[str]:
keys = _active_agent_keys()
found: list[str] = []
for m in MENTION_RE.findall(text or ""):
k = normalize_agent_key(m) or m.lower()
if k in keys and k not in found:
found.append(k)
return found
def _history_as_chat(messages: list[dict[str, Any]], *, limit: int = 20) -> list[dict[str, str]]:
out: list[dict[str, str]] = []
for m in messages[-limit:]:
st = m.get("sender_type")
content = (m.get("content") or "").strip()
if not content:
continue
if st == "user":
name = m.get("username") or "User"
out.append({"role": "user", "content": f"{name}: {content}"})
elif st == "agent":
key = m.get("agent_key") or "agent"
out.append({"role": "assistant", "content": f"@{key}: {content}"})
else:
out.append({"role": "system", "content": content})
return out
async def _hermes_reply(
messages: list[dict[str, str]],
*,
hermes_user: str,
timeout: float | None = None,
) -> tuple[str, dict[str, Any]]:
"""Force Hermes/mimo — never DeepSeek/OpenRouter fallback."""
prov = llm_router.resolve_provider(None)
ptype = (prov.get("provider_type") or "").lower()
if ptype != "hermes":
hermes = llm_router._env_hermes_provider()
if not hermes:
raise RuntimeError("Hermes/mimo niet geconfigureerd (HERMES_API_KEY / :8642).")
raise RuntimeError(
f"War Room vereist Hermes/mimo, huidige provider is '{ptype or 'onbekend'}'. "
"Zet HERMAN_LLM_BACKEND=hermes."
)
reply, meta = await llm_router.chat_messages(
messages,
timeout=float(timeout if timeout is not None else WARROOM_TIMEOUT),
user=hermes_user,
max_tokens=WARROOM_MAX_TOKENS,
temperature=0.35,
)
meta = dict(meta or {})
if meta.get("provider_type") != "hermes":
raise RuntimeError("Antwoord kwam niet van Hermes/mimo — afgebroken.")
return reply, meta
def _soul_system(soul: dict[str, Any], *, room_kind: str) -> str:
name = soul.get("display_name") or soul.get("agent_key")
key = soul.get("agent_key")
role = soul.get("role_title") or "agent"
soul_md = (soul.get("soul_md") or "")[:900]
resp = (soul.get("responsibilities") or "")[:400]
extra = (
"War Room: antwoord in max 4 korte zinnen (NL). Alleen @mention als écht nodig."
if room_kind == "warroom"
else "1:1 DM: bondig in het Nederlands (max 5 zinnen)."
)
return (
f"Je bent {name} (@{key}), {role} bij Foodlinkk.\n"
f"Focus: {resp or '—'}\n"
f"{soul_md}\n\n{extra}"
)
async def _agent_speak(
room: dict[str, Any],
agent_key: str,
*,
trigger: str,
hermes_user: str,
in_reply_to: str | None = None,
) -> dict[str, Any]:
soul = agent_souls.get_soul(agent_key)
if not soul:
return _insert_message(
int(room["id"]),
sender_type="system",
content=f"Agent @{agent_key} niet gevonden.",
metadata={"error": True},
)
hist = list_messages(int(room["id"]), limit=HISTORY_LIMIT)
messages: list[dict[str, str]] = [
{"role": "system", "content": _soul_system(soul, room_kind=room.get("kind") or "warroom")},
]
messages.extend(_history_as_chat(hist, limit=8))
cue = trigger
if in_reply_to:
cue = f"(Reageer op @{in_reply_to}) {trigger}"
messages.append({"role": "user", "content": cue})
try:
reply, meta = await _hermes_reply(messages, hermes_user=hermes_user)
except Exception as exc:
log.exception("agent speak failed %s", agent_key)
return _insert_message(
int(room["id"]),
sender_type="system",
content=f"@{agent_key}: Hermes/mimo fout — {exc}",
metadata={"error": True, "agent_key": agent_key},
)
return _insert_message(
int(room["id"]),
sender_type="agent",
agent_key=agent_key,
content=reply.strip() or "(geen antwoord)",
metadata={
"llm": meta,
"in_reply_to": in_reply_to,
"model": "hermes-agent/mimo",
},
)
async def _orchestrate_warroom(
room: dict[str, Any],
user_text: str,
*,
hermes_user: str,
username: str | None,
) -> list[dict[str, Any]]:
"""Herman facilitates; mentioned agents reply; limited A2A rounds."""
produced: list[dict[str, Any]] = []
mentions = parse_mentions(user_text)
speakers = mentions[:]
if not speakers:
# Herman opens, then may suggest others — still let Herman speak first
speakers = ["herman"]
# Round 0: initial speakers in parallel (sneller bij meerdere @mentions)
keys = speakers[:MAX_REPLIES_PER_ROUND]
if len(keys) == 1:
produced.append(
await _agent_speak(room, keys[0], trigger=user_text, hermes_user=hermes_user)
)
elif keys:
produced.extend(
await asyncio.gather(
*[_agent_speak(room, k, trigger=user_text, hermes_user=hermes_user) for k in keys]
)
)
# A2A follow-up rounds from @mentions in agent replies
for _round in range(MAX_ROUNDS - 1):
next_speakers: list[tuple[str, str]] = [] # (agent, replied_to)
seen = {m.get("agent_key") for m in produced if m.get("sender_type") == "agent"}
for m in produced[-MAX_REPLIES_PER_ROUND:]:
if m.get("sender_type") != "agent":
continue
from_key = m.get("agent_key") or ""
for target in parse_mentions(m.get("content") or ""):
if target == from_key:
continue
if target in seen and (target, from_key) in [(a, b) for a, b in next_speakers]:
continue
next_speakers.append((target, from_key))
# unique preserve order
uniq: list[tuple[str, str]] = []
used: set[str] = set()
for t, src in next_speakers:
if t in used:
continue
used.add(t)
uniq.append((t, src))
uniq = uniq[:MAX_REPLIES_PER_ROUND]
if not uniq:
break
for target, src in uniq:
msg = await _agent_speak(
room,
target,
trigger=(
f"War Room topic van {username or 'user'}: {user_text}\n"
f"@{src} noemde je. Reageer kort en concreet."
),
hermes_user=hermes_user,
in_reply_to=src,
)
produced.append(msg)
return produced
def post_system_line(content: str, *, agent_key: str | None = None, metadata: dict[str, Any] | None = None) -> dict[str, Any]:
"""Proactive system/agent bubble into #war-room (no LLM)."""
room = ensure_warroom()
return _insert_message(
int(room["id"]),
sender_type="system" if not agent_key else "agent",
agent_key=agent_key,
content=content[:2000],
metadata=metadata or {"proactive": True},
)
def announce_handoff(from_agent: str, to_agent: str, handoff_type: str = "partner", *, correlation_id: str | None = None) -> None:
"""Best-effort War Room line for scheduler/API handoffs."""
try:
src = (from_agent or "").strip()
dst = (to_agent or "").strip()
if not src or not dst:
return
label = (handoff_type or "partner").strip() or "partner"
text = f"Handoff: {src} → {dst} ({label})"
post_system_line(
text,
agent_key=src,
metadata={
"proactive": True,
"kind": "handoff",
"from_agent": src,
"to_agent": dst,
"handoff_type": label,
"correlation_id": correlation_id,
},
)
except Exception:
log.exception("warroom announce_handoff failed")
def _normalize_attachments(attachments: list[dict[str, Any]] | None) -> list[dict[str, Any]]:
out: list[dict[str, Any]] = []
for a in attachments or []:
if not isinstance(a, dict):
continue
url = str(a.get("url") or "").strip()
if not url.startswith("/static/uploads/warroom/"):
continue
out.append({
"url": url,
"name": str(a.get("name") or "bestand")[:180],
"mime": str(a.get("mime") or "application/octet-stream")[:120],
"size": int(a.get("size") or 0),
})
if len(out) >= 5:
break
return out
def _trigger_text(text: str, attachments: list[dict[str, Any]]) -> str:
if not attachments:
return text
names = ", ".join(a.get("name") or "bestand" for a in attachments)
if text:
return f"{text}\n\n[Bijlage(n): {names}]"
return f"[Deelde bestand(en): {names}]"
async def post_user_message(
room_id: int,
content: str,
*,
username: str | None = None,
role: str | None = None,
sender_avatar_url: str | None = None,
attachments: list[dict[str, Any]] | None = None,
) -> dict[str, Any]:
text = (content or "").strip()
files = _normalize_attachments(attachments)
if not text and not files:
raise ValueError("Leeg bericht")
room = get_room(room_id)
if not room:
raise ValueError("Room niet gevonden")
hermes_user = hermes_memory.resolve_hermes_user(username=username, role=role, channel="warroom")
meta: dict[str, Any] = {"hermes_user": hermes_user}
if files:
meta["attachments"] = files
user_msg = _insert_message(
room_id,
sender_type="user",
content=text or ("📎 " + ", ".join(a["name"] for a in files)),
username=username,
sender_avatar_url=sender_avatar_url,
metadata=meta,
)
trigger = _trigger_text(text, files)
agent_msgs: list[dict[str, Any]] = []
kind = room.get("kind")
try:
if kind == "dm":
key = room.get("agent_key")
if not key:
raise ValueError("DM zonder agent")
agent_msgs.append(
await _agent_speak(room, str(key), trigger=trigger, hermes_user=hermes_user)
)
else:
agent_msgs = await _orchestrate_warroom(
room, trigger, hermes_user=hermes_user, username=username
)
except Exception as exc:
log.exception("warroom post failed")
agent_msgs.append(
_insert_message(
room_id,
sender_type="system",
content=f"Hermes/mimo fout: {exc}",
metadata={"error": True},
)
)
return {
"ok": True,
"room": room,
"user_message": user_msg,
"messages": [user_msg, *agent_msgs],
"hermes_user": hermes_user,
}
async def _agent_speak_stream(
room: dict[str, Any],
agent_key: str,
*,
trigger: str,
hermes_user: str,
in_reply_to: str | None = None,
):
"""Yield ('start'|'token'|'done'|'error', payload)."""
soul = agent_souls.get_soul(agent_key)
if not soul:
msg = _insert_message(
int(room["id"]),
sender_type="system",
content=f"Agent @{agent_key} niet gevonden.",
metadata={"error": True},
)
yield ("error", msg)
return
hist = list_messages(int(room["id"]), limit=HISTORY_LIMIT)
messages: list[dict[str, str]] = [
{"role": "system", "content": _soul_system(soul, room_kind=room.get("kind") or "warroom")},
]
messages.extend(_history_as_chat(hist, limit=8))
cue = trigger
if in_reply_to:
cue = f"(Reageer op @{in_reply_to}) {trigger}"
messages.append({"role": "user", "content": cue})
yield ("start", {"agent_key": agent_key, "display_name": soul.get("display_name") or agent_key})
chunks: list[str] = []
try:
async for delta in llm_router.stream_chat_messages(
messages,
timeout=WARROOM_TIMEOUT,
user=hermes_user,
max_tokens=WARROOM_MAX_TOKENS,
temperature=0.35,
):
chunks.append(delta)
yield ("token", {"agent_key": agent_key, "delta": delta})
reply = "".join(chunks).strip() or "(geen antwoord)"
msg = _insert_message(
int(room["id"]),
sender_type="agent",
agent_key=agent_key,
content=reply,
metadata={
"llm": {"provider_type": "hermes", "hermes_user": hermes_user, "streamed": True},
"in_reply_to": in_reply_to,
"model": "hermes-agent/mimo",
},
)
yield ("done", msg)
except Exception as exc:
log.exception("agent stream failed %s", agent_key)
msg = _insert_message(
int(room["id"]),
sender_type="system",
content=f"@{agent_key}: Hermes/mimo fout — {exc}",
metadata={"error": True, "agent_key": agent_key},
)
yield ("error", msg)
async def post_user_message_stream(
room_id: int,
content: str,
*,
username: str | None = None,
role: str | None = None,
sender_avatar_url: str | None = None,
attachments: list[dict[str, Any]] | None = None,
):
"""Async generator of SSE-ready event dicts for live typing."""
text = (content or "").strip()
files = _normalize_attachments(attachments)
if not text and not files:
raise ValueError("Leeg bericht")
room = get_room(room_id)
if not room:
raise ValueError("Room niet gevonden")
hermes_user = hermes_memory.resolve_hermes_user(username=username, role=role, channel="warroom")
meta: dict[str, Any] = {"hermes_user": hermes_user}
if files:
meta["attachments"] = files
user_msg = _insert_message(
room_id,
sender_type="user",
content=text or ("📎 " + ", ".join(a["name"] for a in files)),
username=username,
sender_avatar_url=sender_avatar_url,
metadata=meta,
)
yield {"event": "user", "message": user_msg}
trigger = _trigger_text(text, files)
kind = room.get("kind")
speakers: list[str] = []
if kind == "dm":
key = room.get("agent_key")
if key:
speakers = [str(key)]
else:
speakers = parse_mentions(trigger) or ["herman"]
speakers = speakers[:MAX_REPLIES_PER_ROUND]
for key in speakers:
async for kind_ev, payload in _agent_speak_stream(
room, key, trigger=trigger, hermes_user=hermes_user
):
if kind_ev == "start":
yield {"event": "typing_start", **payload}
elif kind_ev == "token":
yield {"event": "typing", **payload}
elif kind_ev == "done":
yield {"event": "agent", "message": payload}
elif kind_ev == "error":
yield {"event": "system", "message": payload}
# light A2A: one follow-up round max from last agent reply mentions
# (non-streamed for simplicity / speed)
yield {"event": "done", "ok": True}