DCS / AI & Workspace / Assistant
General-purpose · cited · signed

An assistant that cites its sources — and signs them.

DCS Assistant researches across the web + your business memory, drafts long-form, runs analysis, calls your tools. Every claim cites a source. Every output emits a receipt. The "I made this up" problem is structurally eliminated.

DCS Assistant How do I deploy an agent? user · 2m ago DCS ASSISTANT To deploy an agent, run: $ dcs deploy --type=agent This will spin up a worker in ~30s. You can monitor it via the dashboard Ask anything about your DCS workspace…
Capabilities

One assistant, six modes of work.

🔍

Deep research

Multi-source web research with citations. Cross-checks claims. Flags conflicts. Returns a sourced brief.

✍️

Long-form drafting

Memos, reports, proposals, blog posts. Adapts to your brand voice from your DCS Memory.

📊

Data analysis

Drop a CSV/Excel. Get statistical summaries, charts, anomalies. Optionally runs Python sandbox.

🛠️

Tool use

Call your APIs, query Postgres, hit Stripe, post to Slack — all via plain English instructions.

📅

Calendar + email triage

"Find a slot next week with both Maria and Tom." "Summarize my unread emails into action items."

💡

Knowledge from Memory

Remembers your decisions, your customers, your team. Every fact references a Memory CID for audit.

Deep research

The "did you actually check?" problem, solved.

Most AI gives plausible answers. Assistant gives sourced answers — every factual claim links to the URL it came from, and the receipt chain proves Assistant actually fetched and read the source, not invented it.

Plans the query

Decomposes your question into 3-7 sub-questions. Shows the plan so you can edit before it runs.

Browses real-time

Hits live web with our search agent. Reads full pages (not just snippets). Stores each page hash in the receipt.

Cross-checks

If two sources disagree, Assistant flags it: "vendor X says Y, regulator Z says W — here's both."

Tool use

Plain English in. Real actions out.

Connect any of 10+ pre-built connectors or define your own. Assistant translates intent into API calls — every call signed and audit-able.

💳

Stripe

"Refund order #4182" → fetches order, asks confirm, runs refund, replies with txid.

🗃️

Postgres / Supabase

"How many bookings last week?" → safe read-only query, returns count + chart.

📧

Gmail / Outlook

"Draft a reply to the supplier about late delivery." → drafted, you approve before send.

📅

Calendar

"Schedule a 30-min call with Tom for next Tue afternoon." → checks both calendars, sends invite.

Pricing

Usage-based. Free for light use.

Free

$0 · 100 messages/month

Tier 0 inference, basic tools, 2 connectors.

Pro

$29/mo · 5,000 messages

Frontier models on demand, all connectors, deep research, voice.

Sovereign

Talk · BYO weights

Assistant runs on your inference + your Memory inside your VPC.

FAQ

Common questions.

Why does this exist alongside ChatGPT / Claude?
DCS Assistant is opinionated — it cites sources, signs receipts, calls *your* tools, reads *your* memory. ChatGPT/Claude are general-purpose chatbots. Assistant is a teammate inside your DCS stack.
Can it hallucinate?
Lower risk by design — Assistant is constrained to cite sources for factual claims. If it can't find a source, it says "I couldn't verify this." For tool calls, every action requires either a confirmation prompt or a signed plan. But zero-hallucination is impossible to claim — we ground heavily, we don't lie about uncertainty.
What models does it run?
Tier 0: open-weight on DCS H200 (Llama, Qwen, Mistral, DeepSeek). Tier 1: frontier passthrough (Claude, GPT, Gemini). Tier 2: your own vLLM endpoint. Assistant picks model per task — research uses frontier, summarization uses Tier 0.
Does it learn from my data?
Yes, into your DCS Memory only. We don't train foundation models on your data. Your Memory is yours — export, delete, audit any time.
How is this different from DCS Edit?
Edit modifies your DCS-built site. Assistant is a general work assistant — research, drafting, analysis, tool use. Different scope, same receipt chain.

An assistant that cites its work.

Stop pasting "is this true?" into Google after every AI answer.

Open Assistant → See DCS Memory