Coder

MemStrata Coder - Local Context for AI Coding Tools

Route focused project context into coding tools and measure each visible user turn.

Who it is for: Developers and teams using AI coding tools who need current project context without sending a whole repository on every prompt.

Install Coder Compare products

Five-minute flow

01

Install

Install the compiled Coder package from PyPI on Linux x86_64, then run memstrata-coder init.
02

Connect a tool

Use a supported editor or terminal path. Unsupported clients are labelled rather than implied.
03

Measure a turn

Each visible user prompt is counted once. The dashboard separates input, cache, and output.

Context receipt for a visible user turn

Coder records what was retrieved, what was sent, and how that compares with a session-start counterfactual on the first eligible invocation. Auxiliary model calls are excluded from the input-savings total.

evidence receipt

product: coder

output: Route focused project context into coding tools and measure each visible user turn.

proof: visible on device

Controlled result

81.4% input-token avoidance observed in a four-turn controlled Coder E2E test against a one-time full-repository session-start counterfactual. Results vary by repository, client, model, and task.

Sample: Four eligible completed visible user turns across four distinct test sessions in a controlled Coder E2E database. Not a production cohort.

Sample is four eligible visible user turns, not a cross-user median.
Baseline is a one-time full-repository session-start counterfactual, not naive RAG.
A visible user prompt is counted once even when a client makes multiple model invocations.
The full-repository counterfactual is assigned only to the first eligible invocation of a conversation.
Auxiliary invocations are excluded from the input-savings total.
Saved input is max(counterfactual_input_tokens - actual_input_tokens, 0).
Results vary by repository, client, model, and task.

Methodology

Capabilities

Project discovery and focused context instead of whole-repo dumps
VS Code extension plus VS Code shims (Cursor, Windsurf, Continue, Cline, Copilot Chat)
IDEs: VS Code, JetBrains AI, Zed
Terminal agents: Claude Code, Codex CLI, Aider, OpenCode, Claude Desktop
Visible-user-turn telemetry distinct from model-invocation counts

Works in IDEs, VS Code shims, and the terminal

Coder connects through MCP, an OpenAI-compatible harness, or the VS Code extension. The extension also loads in VS Code shims such as Cursor and Windsurf. Native tab-complete and some closed Composer paths are partial, and we label them that way.

IDEs

Install once, then connect the editor you already use.

VS Code · MCP + Extension
JetBrains IDEs · MCP
Zed · MCP + Harness

VS Code and VS Code shims

The VS Code extension runs in VS Code and in forks that load VS Code extensions, including Cursor and Windsurf.

Cursor · MCP + Extension
Windsurf · MCP + Extension
Continue.dev · All three
Cline · MCP + Harness
GitHub Copilot Chat · MCP + Extension

Terminal and CLI agents

Point MCP or an OpenAI-compatible harness at the local Coder runtime.

Claude Code · MCP + Harness
Codex CLI · Harness
Aider · Harness
OpenCode · MCP + Harness
Claude Desktop · MCP

Privacy and data flow

Privacy-critical content and indexes remain on the device by default. OAuth, signed entitlements, device registration, and bounded private decisions use MemStrata services. A user-selected cloud model or connector receives only the data required for that explicit operation. Credentials are stored in the operating-system credential vault, not in page content or analytics. Customer content is not used for training unless the customer separately opts in.

Feature	Local data	Possible outbound	Trigger	Credential location
OAuth, signed entitlements, and device registration	Device and entitlement cache	Identity and device-bound entitlement fields	Sign-in or refresh	Operating-system credential vault
User-selected cloud model or connector	Local evidence selection and indexes	Only the data required for that explicit operation	User configures and sends the request	Operating-system credential vault
MemStrata-managed email	Local alert or report draft	Minimized message and recipient	User-configured delivery; one message per signed-in user per UTC day during trial	Server secret plus entitlement
Project indexing and context routing	Repository indexes, context receipts, and local telemetry	Selected context to a user-chosen coding model or MCP client	Editor or terminal request after the user connects a tool	Operating-system credential vault for model keys

Install and trial

Use the complete product free for 180 days. No subscription is needed during the trial. A paid entitlement is required only after the trial ends.

The 180-day trial begins after successful product activation. Product functionality remains enabled during the trial. MemStrata-managed email is limited to one message per signed-in user per UTC day. Reinstalling does not restart access. A paid entitlement is required after the trial.

PyPI · Linux only · 0.6.32

python -m pip install memstrata-coder

sha256 e3c3a5c39a576963f76a000bdcf455452af051f0371c4b76645a43e1166479a5

Current latest PyPI file is Linux only.

microsoft store

Coming soon
mac app-store

Coming soon
snap store

Coming soon

Integrations: VS Code, Cursor, Windsurf, JetBrains IDEs, Zed, Continue.dev, Cline, GitHub Copilot Chat, Claude Code, Codex CLI, Aider, OpenCode, Claude Desktop

FAQ

Does Coder require a specific model?

No. You select the coding model. Local model results are not treated as monetary savings when no paid model cost applies.

Is the 81.4% figure a typical user result?

No. It is a four-turn controlled E2E result with a full-repository session-start counterfactual. It is not a median, average, or guarantee.

Install Coder