SaMo Chat

SaMo Chat is a general-purpose AI assistant in Visual Studio Code for everyday questions and software development across programming languages, frameworks, and project types. Understand codebases, implement changes, investigate issues, run regression checks, and review work from one integrated workspace.

Designed for individual developers and teams who want a focused way to move from engineering task to verified change.

Product capabilities

• AI coding workflow: ask questions, understand a project, develop features, debug, audit, and review code from the SaMo Activity Bar. SaMo handles root-cause debugging, end-to-end features, multi-file and repository-wide changes, codebase exploration, tests and build failures, code review, documentation, API/database integrations, performance work, UI changes, and framework migrations. Edit requests are applied to project files and offered with Keep/Undo controls; explanation, suggestion, and review requests stay read-only unless you ask for changes.
• Iterative agent tools: for coding work, SaMo can search and read workspace files, inspect definitions/references/call hierarchy/types and diagnostics, examine Git state, apply validated edits, and record verification evidence. The loop is bounded; it does not guarantee a particular response latency or correctness rate, which depend on the SaMo service and model.
• Controlled code changes: review proposed edits and choose to keep or undo them. SaMo uses SHA-256 preconditions to prevent stale overwrites, blocks protected paths and package-manager lockfiles, and keeps changes inside the workspace or a directory you explicitly authorized.
• On-demand local access: when workspace evidence shows an explicitly authorized additional directory is needed, the model can request it. You select the folder and confirm the grant; SaMo then adds its scoped file tools to the active agent loop. A denied or cancelled request cannot be prompted again during that task, and sensitive paths remain blocked.
• Regression checks: when samo.runTests is enabled, the model can choose to detect and request tests, type checks, lint, and build checks. Each command requires individual approval. SaMo reports only checks confirmed by actual tool results. Stop cancels active model requests and best-effort stops running commands.
• Conversation continuity: keep up to 50 conversations per workspace, start a fresh conversation, or return to saved work. Existing chat history is migrated automatically.
• GitHub workflows: review Git diffs and create pull requests using the GitHub CLI and the credentials already configured in your environment.
• MCP integrations (opt-in): connect user-configured stdio MCP servers in trusted workspaces. SaMo asks before starting each server and before every individual tool call, including calls selected by the bounded coding-agent loop.
• Developer context: optionally include the active selection or file (up to 12,000 characters). Code-fix requests include editor context automatically when an editor is open.
• Account-based access: connect to the SaMo VS Portal with an access token stored in VS Code's encrypted SecretStorage. The portal meters usage and routes requests to its configured model provider.
• Task-aware inference: casual conversation and code-generation requests use separate model sampling settings, while the portal applies SaMo's behavior, honesty, and security policy.
• Responsive request handling: chat answers stream as they are generated, independent read-only tools can run concurrently, and the agent enforces a total runtime budget (180 seconds by default; configurable up to 480 seconds). The activity indicator shows the current stage and elapsed time, with a longer-wait hint and a Cancel control for slow requests. The coding agent avoids resending chat history already represented in its task prompt. Request timing and payload-size metrics (headers, first token, total duration, request bytes, and response characters) are available in View → Output → SaMo without logging prompt or answer text. Exact input/output token counts and model selection require usage and routing support from the SaMo portal. Connection and HTTP errors include actionable details and a user-triggered retry. Individual API calls time out after 120 seconds without response data by default; configure samo.requestTimeoutSeconds from 15 to 300 seconds.

Plans

Creating an account is free, but it does not include model requests. Paid plans are one-time purchases for 30 days and do not renew automatically. Usage credits are cost-weighted, and server-side inference budgets cap estimated model spend. Check the SaMo VS Portal for current availability and account-specific plan information.

Enterprise plans can be tailored for team or business requirements. Contact info@samotech.co.in to discuss your needs.

Get started

1. Create an account at vs.samotech.co.in and copy the access token from your dashboard.
2. In VS Code, run SaMo: Set Access Token and paste the token. It is stored in encrypted SecretStorage, not in settings.
3. Set samo.email to the email address associated with your portal account. The default API endpoint is https://vs.samotech.co.in; change samo.apiUrl only if you use another SaMo-compatible service.
4. Open the SaMo icon in the Activity Bar and start working. Use SaMo: Open Account Portal to manage your account, usage, and plan.

Commands

MCP server setup

Configure samo.mcp.servers in VS Code User Settings (not workspace settings), for example:

"samo.mcp.servers": [
  {
    "name": "example",
    "command": "npx",
    "args": ["-y", "some-mcp-server"]
  }
]

Use a trusted workspace. SaMo prompts before launching a configured executable and asks separately before each tool invocation. During coding tasks, SaMo can discover and propose configured MCP tools through its bounded agent loop; each proposed call still requires approval. Credentials should come from the server process's inherited environment rather than being placed in user settings; the extension passes only VS Code's safe inherited environment variables plus explicitly configured overrides. Tool arguments and results are not added to saved chat history.

Privacy

SaMo sends your message, conversation history, and configured account email to the selected SaMo API endpoint (the SaMo VS Portal by default). When editor context is included, the selected code or active file content (up to 12,000 characters), workspace-relative path, and language identifier are also sent. Code-fix requests include available editor context automatically. Screenshot attachments are processed locally using OCR; only extracted text is sent to the API. The portal forwards chat content to its configured model provider. MCP server tool inputs and results stay local to the extension and server connection; they are not sent to the hosted model by the current manual command flow. Do not submit credentials, secrets, or code you are not authorized to share. Review the portal's privacy information before use.

Business and support

For account, plan, team, or product support, email info@samotech.co.in. Include your SaMo account email and a brief description of the issue; do not include access tokens, passwords, or private source code.

Experimental model fine-tuning

The optional training toolkit contains a small behavior dataset, evaluation prompts, and a GPU LoRA starter. It is not shipped in the VS Code package, and SaMo has not been fine-tuned. The current hosted portal model is steered by an inference-time system prompt; creating a LoRA adapter requires training and deploying compatible model weights separately.

Publishing (maintainer notes)

npm install
npm run build
npx vsce package   # produces samo-chat-<version>.vsix
npx vsce login devsanjib
npx vsce publish