Deployment & data flow

Your infrastructure.
Explicit boundaries.

Engram’s index and services run in your environment. Configure source access, model providers and repository authorization to match how your team works.

Where the data goes.

Source systems feed the self-hosted Engram index. People and AI clients query it. Configured embedding and reasoning providers receive selected text.
Connected sources

Code & work history

Repositories, commits, work items and review evidence.

Your deployment

Engram services & index

Parse, resolve, embed and compile. Store source context, vectors, graph relationships and retained history in your database.

Your users & tools

MCP, CLI & Explore

Return context to authenticated clients according to the access controls you configure.

Configured model endpoints · local or external

Embedding: source chunks and search text go to the configured embedding service.

Optional reasoning: selected context and prompts go to the model used by enabled features or your AI client.

A local Ollama embedding endpoint is an option. It does not make separately configured reasoning features or AI clients offline. Review each enabled path and its provider settings.

Configure access before connecting users.

Identity and repository scope

Repository authorization supports Disabled, diagnostic Shadow and Enforce modes. The configuration default is Disabled. Select enforcement deliberately, map authenticated identities to repository access, and validate allowed and denied queries.

Confirm identity propagation for each surface you enable. Optional bot integrations need a separate scope review; do not assume a client’s access model applies to every integration.

Network and credentials

Provide access to your source and work-item APIs, the PostgreSQL index, configured model endpoints and any enabled integrations. Release installation also needs access to its distribution artifacts.

Keep source tokens and model credentials in your environment’s secret-management facilities. Use TLS and restrict service exposure to the clients and operators that need it.

Plan the operating path.

Versioned installation

Docker Compose and OCI Helm packaging provide versioned deployment paths. Confirm storage, ingress, resources, authentication and model settings against your chosen release. Enterprise chart customization continues to evolve.

Indexing and freshness

Use full, incremental or artifact-recompile jobs as appropriate. Unchanged work can be reused; graph and cross-file processing vary by source provider. Monitor jobs and indexed revisions rather than assuming every successful query is current.

Recovery and upgrades

Backup, export and restore workflows support recovery planning. Test a restore using your release and data. Reverting an application image does not reverse database migrations; upgrade and recovery compatibility must be checked.

Offline license verification is available when configured. A complete air-gap distribution bundle is an exploratory roadmap item. Release packaging does not imply a compliance certification or universal infrastructure compatibility.

Coverage you can evaluate

14 languages. Different depths of understanding.

This table describes symbol extraction foundations. Call resolution, imports, contracts, framework conventions and authored-rationale extraction have separate limits.

LanguagesExtraction foundationEvaluate on your code
C#Bundled tree-sitter AST grammarStatic resolution, dependency injection and framework dispatch can differ from runtime behavior.
TypeScript · JavaScriptBundled tree-sitter AST grammarsCheck dynamic calls, module boundaries and framework-specific conventions.
PythonBundled tree-sitter AST grammarImport-aware resolution has known gaps; check package layouts and dynamic dispatch.
Java · Go · Rust · RubyBundled tree-sitter AST grammarsCheck language-specific call resolution and the frameworks used in your estate.
PHPBundled tree-sitter AST grammarTrait handling and deeper resolution remain areas of improvement.
C · C++Bundled tree-sitter AST grammarsValidate macro-heavy code, include layouts and indirect calls.
SQLCustom tree-sitter grammar, with ScriptDom for T-SQL analysisDialect and declaration forms affect coverage, including function variants.
Kotlin · SwiftRegex fallback for symbol extractionShallower structural extraction while AST bindings and language support evolve.

Use a representative repository and verify source locations, missing symbols and incorrect edges. The public example shows why those checks matter.

Before a pilot

Agree the install
and the evidence.

  1. Select representative repositories and the history providers to connect.
  2. Choose deployment and model endpoints; confirm credentials, network access and repository enforcement.
  3. Use the installation instructions and artifacts for the agreed release. Verify readiness and one allowed / denied access case.
  4. Record the indexed revision, run the evaluation tasks, and review gaps with the team.

Request the current release instructions during scoping. The installation and support scope is confirmed in your proposal.

Use the evaluation worksheet →
See Engram in context

Bring a real codebase question.

We’ll use a 30-minute scoping call to understand your repositories, tooling and evaluation goals.

Book a scoping call