TAKpilot

AI Chat Copilot for CloudTAK

A Claude-powered chat copilot embedded directly inside CloudTAK. Command your Common Operating Picture through natural language — place typed markers, brief missions, and manage your COP without clicking through menus.

CloudTAK-native AGPL-3.0 open source Model-agnostic Edge-capable MIL-STD-2525 / APP-6

Platform Demo

Key Features

Chat-Native COP Control

Place markers, list missions, build data packages, and subscribe to channels through a single chat interface embedded in CloudTAK — no menu navigation required.

Vision & File Intelligence

Drop PNG, JPG, or PDF files directly into the chat. The agent reads them using built-in vision — no separate OCR pipeline. Extract entities from SITREPs and sketches, then place them on the map.

Streaming Tool Cards

Every tool call streams to the UI as a collapsible card showing name, inputs, status, timing, and result preview. Full audit trail of what the agent did — and when.

Approve / Deny Control

Destructive operations — marker create, mission write, channel subscribe — never execute silently. Each triggers an Approve / Deny prompt showing the resolved action, coordinates, and NATO symbol before anything is written.

Model-Agnostic & Edge-Capable

Switch between Claude Opus, Sonnet, and Haiku per session. Edge-capable open models (Llama 3.3 70B, Qwen 2.5 72B, Mistral Large) can be served on operator-owned hardware for low-connectivity and air-gapped environments.

MIL-STD-2525 / APP-6 Symbology

Confirmation prompts and tool cards render the correct NATO / APP-6 icon next to every type code. The operator sees the actual military symbol before approving placement — not just the cryptic CoT string.

How It Works

TAKpilot operates as a chat panel inside CloudTAK, backed by an agent that runs in a secure per-session sandbox. Every action the agent takes is visible, auditable, and gated by the operator.

01

Operator sends a message or drops a file

Type a question, issue a command, or drag-drop a SITREP PDF or recon sketch. Files are written to a private session sandbox — no cross-session access, never stored permanently.

02

Agent reads, reasons, and plans tool calls

The AI model reads any attached files using built-in vision, queries CloudTAK to gather context, and plans the minimum set of tool calls needed to fulfil the request.

03

Tool calls stream as cards; destructive ones pause for approval

Each tool call appears as a live card. Read-only calls run automatically. Any write — marker, mission, package — triggers an Approve / Deny prompt with full details and the NATO icon before execution.

04

Agent acts as the operator's identity

Approved calls hit CloudTAK's REST API using the operator's session token. Every write is attributed to the operator in CloudTAK's audit log — no service account, no shared identity.

Challenges it solves

  • Operators lose time navigating menus to place markers, check missions, and update layers under time pressure
  • Unstructured intel — handwritten sketches, SITREP PDFs, image overlays — has no fast path onto the COP
  • Tool calls and AI-driven actions are invisible, making it impossible to audit what the system did or why
  • Destructive AI actions (delete, overwrite, broadcast) execute without an opportunity to review or refuse
  • Model lock-in forces operators onto a single provider even when the network or threat model demands otherwise

Value

  • Faster COP updates — a single sentence replaces a multi-step menu workflow
  • Unstructured data becomes typed, georeferenced COP objects in one drag-drop operation
  • Full transparency — every tool call is shown, timed, and logged with the operator's identity
  • Zero-surprise writes — the operator approves every destructive action with full context before it executes
  • Deploy frontier cloud or edge-local models based on connectivity and operational security requirements

Target Audience

Tactical operators on CloudTAK Mission commanders and S3 staff ISR and reconnaissance teams Intelligence analysts (S2) Ukrainian Defense Forces NATO partner units on TAK Public safety and emergency management Defense integrators and CloudTAK deployers

Distribution

TAKpilot is available as an open-source package and through the Brave1 defense innovation marketplace for Ukrainian and NATO partner organisations.

Related Services

TAKpilot is built on Corvus Intelligence's custom software development capabilities. Explore the engineering services behind the platform.

Frequently Asked Questions

What is TAKpilot?

TAKpilot is an AI chat copilot embedded directly inside CloudTAK. It allows tactical operators to command their Common Operating Picture through natural language — placing MIL-STD-2525 / APP-6 typed markers, briefing missions, discovering channels, and building data packages, all without navigating menus.

Who controls the data? Does my operational data leave my network?

No operational data is stored by TAKpilot. All files uploaded to the chat are processed in a per-session sandbox directory and deleted when the session ends. The only external call is to the AI model provider (Anthropic or an on-network Bedrock / Vertex endpoint). Deployers can configure TAKpilot to use AWS Bedrock or Google Vertex to keep model calls inside an existing cloud enclave.

Can TAKpilot run offline or on air-gapped networks?

TAKpilot requires a connection to an AI model endpoint. Cloud frontier models require internet access to Anthropic's API. For air-gapped or low-connectivity deployments, TAKpilot supports edge-capable open models (Llama 3.3 70B, Qwen 2.5 72B, Mistral Large) served on operator-owned hardware via an OpenAI-compatible endpoint.

What AI models does TAKpilot support?

TAKpilot supports Claude Opus 4.7, Sonnet 4.6, and Haiku 4.5 from Anthropic, as well as edge-capable open models including Llama 3.3 70B, Qwen 2.5 72B, and Mistral Large. The operator selects the model per session from a switcher in the chat panel.

What CloudTAK object types can TAKpilot read and write?

TAKpilot reads and writes markers / CoT features, missions and mission logs, mission layers and attached files, data packages, channels, basemaps, and geofences — all as the logged-in operator's identity. Destructive operations require an explicit Approve action before any write is made to CloudTAK.

What are the AGPL-3.0 implications for integrators?

TAKpilot is released under AGPL-3.0, the same licence as CloudTAK upstream. Organisations that deploy TAKpilot internally without distributing it are not required to open-source their modifications. Organisations that distribute a modified version to third parties must make the source available under the same licence.

How does TAKpilot handle destructive actions such as deleting markers?

Every destructive tool call triggers an Approve / Deny prompt in the chat panel before anything is written. The prompt shows the resolved action: coordinates, callsign, MIL-STD-2525 type code, and the corresponding NATO / APP-6 icon. The operator must click Approve; denying cancels the call and the agent receives the denial cleanly.

How is TAKpilot installed on an existing CloudTAK server?

TAKpilot is deployed as a Node.js service alongside your existing CloudTAK instance. Installation requires Node.js 20+, an Anthropic API key (or a configured Bedrock / Vertex / local endpoint), and network access from the TAKpilot service to the CloudTAK REST API. The full step-by-step installation guide is available on GitHub at github.com/UA-WCV/takpilot.

What file formats does TAKpilot accept in the chat for AI analysis?

TAKpilot's chat accepts PNG, JPG/JPEG, GIF, and WebP images, as well as PDF documents. Files are processed using the AI model's built-in vision capabilities — no separate OCR pipeline is required. Uploaded files are stored in a private per-session sandbox and permanently deleted when the session ends.

How can I request access or a demonstration of TAKpilot?

You can request a demonstration via the Book a Demo page, or contact us directly at contact@corvusintell.com.

Contact

Interested in TAKpilot for your unit or organisation? Get in touch for a demo or to discuss deployment requirements.

Request a Demo or email contact@corvusintell.com