By Sagar Shankaran, Founder of CallSphere
Gemini Deep Research Max takes 3–10 minutes per query. LangChain's Deep Agents framework handles process isolation, crash recovery, persistent memory. We cover the architecture and the operational reality of multi-minute LLM runs.
Key takeaways
TL;DR — Deep-research agents run for minutes to hours, not seconds. They need persistent state, crash recovery, sub-agent delegation as a first-class concern. LangChain Deep Agents and Google's Deep Research Max API (April 2026) are the production-ready primitives.
A lead agent owns a long-horizon goal. It maintains:
flowchart TD
GOAL[Multi-hour goal] --> LEAD[Lead agent w/ plan + memory]
LEAD -->|spawn| S1[Sub-agent: read source A]
LEAD -->|spawn| S2[Sub-agent: read source B]
LEAD -->|spawn| S3[Sub-agent: code analysis]
S1 -->|report| LEAD
S2 -->|report| LEAD
S3 -->|report| LEAD
LEAD --> CKPT[(Checkpoint store)]
CKPT --> LEAD
LEAD --> WRITE[Long-form write phase]
WRITE --> OUT[Final report]
Skip when: a 30-second answer would do, or your infra can't tolerate a multi-minute job.
CallSphere doesn't run hours-long jobs in the live voice path (that would be insane). Instead, deep-research lives in two backstage workflows:
Both checkpoint to Postgres every 30 seconds. Across 37 agents · 90+ tools · 115+ DB tables · 6 verticals, these are 2 of the agents (lead + spawn-on-demand sub-agents). Pricing: Starter $149 · Growth $499 · Scale $1,499, 14-day trial, 22% affiliate.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent in your browser — 60 seconds, no signup.
from deepagents import DeepAgent, Subagent
researcher = DeepAgent(
model="claude-opus-4-7",
memory_backend="postgres://...",
checkpoint_every=30, # seconds
subagents=[
Subagent(name="reader", model="gpt-4o-mini", tools=[fetch_url]),
Subagent(name="coder", model="gpt-4o", tools=[run_code]),
Subagent(name="searcher", model="gpt-4o-mini", tools=[web_search]),
]
)
result = await researcher.run(
goal="Produce a 30-page strategic brief for vertical: dental practices",
timeout_minutes=30
)
Q: Async vs sync? Always async. Use a job queue and a results-fetch endpoint.
Q: What if a sub-agent fails? Lead retries (max 3) with a different model, then escalates "could not complete sub-goal X" in the final report.
Q: How big can the memory get? Tens of MB of structured notes + a vector store of source chunks. Compact periodically.
Q: Cost? $0.50–$5 per run for typical deep research; depends on model + tool calls.
Q: User-facing UX? Show progress: "step 7 of 12, currently reading PDF X." Don't show a spinner for 25 minutes.
Most write-ups about long-Running Deep-Research Agents stop at the architecture diagram. The interesting part starts when the same workflow has to survive a noisy phone line, a half-typed chat message, and a flaky third-party API on the same day. Once you frame long-running deep-research agents that way, the design choices get easier: short tool descriptions, narrow argument types, and a hard cap on tool calls per turn beat any amount of prompt engineering.
Agentic AI in a real call center is a different beast than a single-LLM chatbot. Instead of one model answering one prompt, you orchestrate a small team: a router that decides intent, specialists that own a vertical (booking, intake, billing, escalation), and tools that read and write to the same Postgres your CRM trusts. Hand-offs are where most production bugs hide — when Agent A passes context to Agent B, anything that isn't explicit in the message gets lost, and the user feels it as the agent "forgetting." That's why the systems that hold up under load are the ones with typed tool schemas, deterministic state stored outside the conversation, and a hard ceiling on tool calls per session. The cost story is just as important: a multi-agent loop can quietly burn 10x the tokens of a single-LLM design if you let it think out loud at every step. The fix isn't a smarter model, it's smaller agents, shorter prompts, cached system messages, and evals that fail the build when p95 latency or per-session cost regresses. CallSphere runs this pattern across 6 verticals in production, and the rule has held every time: the agent you can debug in five minutes will out-survive the agent that's "smarter" on a benchmark.
Still reading? Stop comparing — try CallSphere live.
CallSphere ships complete AI voice agents per industry — 14 tools for healthcare, 10 agents for real estate, 4 specialists for salons. See how it actually handles a call before you book a demo.
Q: What's the hardest part of running long-Running Deep-Research Agents live?
A: Scaling comes from constraint, not capability. The deployments that hold up keep each agent narrow, cap tool calls per turn, cache the system prompt, and pin a smaller model for routing while reserving the larger model for synthesis. CallSphere's stack — 37 agents · 90+ tools · 115+ DB tables · 6 verticals live — is sized that way on purpose.
Q: How do you evaluate long-Running Deep-Research Agents before shipping?
A: Hard ceilings beat heuristics. A maximum step count, an idempotency key on every tool call, and a fallback to a deterministic script when confidence drops below a threshold are what keep the loop bounded. Evals that simulate noisy inputs catch the rest before they reach a real caller.
Q: Which CallSphere verticals already rely on long-Running Deep-Research Agents?
A: It's already in production. Today CallSphere runs this pattern in IT Helpdesk, alongside the other live verticals (Healthcare, Real Estate, Salon, Sales, After-Hours Escalation, IT Helpdesk). The same orchestrator code path serves voice and chat — the difference is the tool set the router exposes.
Want to see salon agents handle real traffic? Spin up a walkthrough at https://salon.callsphere.tech or grab 20 minutes on the calendar: https://calendly.com/sagar-callsphere/new-meeting.
Written by
Sagar Shankaran· Founder, CallSphere
Sagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
Graphiti is the open-source temporal knowledge graph for AI agents in 2026. Learn how bi-temporal memory beats vector RAG for voice agents and long-running LLMs.
How to design a multi-agent system using MCP for tools and A2A for cross-vendor coordination, with a CallSphere voice agent as a participating node.
A2A is the open standard for agent-to-agent coordination. Here is how the Agent Card JSON works, how discovery happens, and what to publish.
A2A unlocks cross-vendor agent coordination, but most enterprise voice/chat workloads still ship faster on a single-vendor stack. Here is how to choose.
Fully autonomous agents are still a fantasy in production. LangGraph's interrupt() lets you pause for human approval mid-graph without losing state. We cover approve/edit/reject/respond actions and CallSphere's escalation ladder.
Neo4j's agent-memory project ships short-term, long-term, and reasoning memory in one graph. Microsoft Agent Framework and LangChain both wire it in. Here is the production pattern.
© 2026 CallSphere LLC. All rights reserved.