The superintelligent context layer between your agents and your codebase. Engram's live code knowledge graph delivers full project context in 167 ms, cuts token usage by ~60%, builds persistent episode memory across sessions, and transforms vague prompts into structured agent instructions — so every session starts smarter than the last.
MCP tools
25
Indexed files
847
Agent memory
∞
dashboard preview
Context stays hot.
mcp tools
25
symbols
12.4k
edges
28.9k
agent memory
∞
recent activity
get_context_for_task — 8 files loaded in 167ms
just now
save_episode — "Used Prisma soft-delete pattern"
2m ago
transform_prompt — bug_fix XML ready for agent
5m ago
recall_episodes — 3 relevant past solutions found
9m ago
mcp config
"engram": {
"command": "engram",
"args": ["serve"]
}language mix
live snapshot“Why does my AI take so long to fix issues in my codebase?”
Because it needs much, much better context.
Less warm-up. More shipping.
Engram acts like shared memory for your coding stack. It keeps your repository indexed, watches changes, and exposes context through 25 focused MCP tools across three tiers of capability.
Persistent graph
Parse once. Incremental updates forever.
Instant switching
Context survives hand-offs and branch changes.
25 MCP tools
Every tool an agent needs — context, memory, orchestration, prompt intelligence.
Error → Fix pipeline
One stack trace gives the agent a complete blast-radius map. Symbols involved, their callers, and recent changes — in one tool call.
get_error_context({
stackTrace: `TypeError: Cannot read 'id'
at UserService.getUser (user.ts:47)
at AuthController.login (auth.ts:23)`
})
// → 3 symbols traced, 7 callers, 2 recent changesPR-aware context
Before an agent touches a branch, it knows what changed, what breaks, who owns the affected code, and which tests cover it.
get_pr_context({
branch: "feat/auth-refactor"
})
// → 14 files changed, 3 owners identified
// → 8 tests cover affected symbolsImpact of change
The #1 thing missing from AI coding — agents make changes without knowing the blast radius. This tool fixes that completely.
get_impact_of_change({
symbolName: "UserModel",
changeDescription: "Add soft-delete field"
})
// → 23 direct dependents, 47 transitive
// → 12 tests need updatingPersistent agent memory
Agents write notes that survive sessions. Future agents in the same project read them. Cross-agent, cross-session memory baked in.
remember({
key: "soft-deletes",
value: "UserService uses soft deletes.
Never call DELETE on users table."
})
// Stored. All future agents can recall this.The full toolkit.
Seven capabilities that give agents and developers superpowers. From multi-repo graphs to test coverage, deep research to task orchestration.
LSP integration
Compiler-accurate context from go-to-definition, find-references, and type info. Agents get type-safe intelligence, not regex guessing.
Test coverage overlay
Which symbols have zero tests? get_uncovered_symbols() returns a prioritized list so agents know what to test next.
explain_architecture
Auto-generates a living architecture doc from the graph. Layers, modules, dependency flow — never stale because it's computed from real code.
Diff-aware re-indexing
After git pull, agents start with a delta: 'These 14 symbols changed since your last session.' No more blank-slate re-exploration.
Engram Replay
Record every agent query and the context returned. Replay to debug why an agent made a bad decision. Invaluable for teams.
Multi-repo graph
One Engram instance spans multiple repos. Cross-repo call graphs and unified architecture views for microservice teams.
Episode Memory
Agents remember how they solved problems. save_episode stores decisions, technologies, and outcomes. recall_episodes surfaces relevant past solutions at session start.
Smart Prompt Transform
transform_prompt converts vague natural language into structured XML with task classification, extracted symbols, constraints, and required output format.
Deep Research
deep_research aggregates codebase context, past episodes, agent notes, and architecture overview into one token-budgeted response. Accurate context, always.
Task Orchestration
decompose_task breaks complex goals into typed subtasks with dependency ordering and suggested agent types. Plan before you code.
Test Coverage
get_test_coverage gives project-wide or per-file coverage stats with configurable thresholds. Flag under-covered files before shipping.
Agent Profile
get_agent_profile builds an inferred preference map from past episodes — preferred libraries, coding conventions, recurring patterns — automatically.
Engram in your editor.
A sidebar that shows the symbol graph, recent changes, who owns what, and architecture — without needing an AI agent at all. Makes Engram useful to every developer on the team.
dependency graph
agent memory
UserService uses soft deletes
JWT with refresh token rotation
A dashboard that feels like instrumentation.
The signed-in experience now reads like a control room: live health, ingest activity, language mix, token management, billing, and team state in one cohesive shell.
dashboard preview
Context stays hot.
mcp tools
25
symbols
12.4k
edges
28.9k
agent memory
∞
recent activity
get_context_for_task — 8 files loaded in 167ms
just now
save_episode — "Used Prisma soft-delete pattern"
2m ago
transform_prompt — bug_fix XML ready for agent
5m ago
recall_episodes — 3 relevant past solutions found
9m ago
mcp config
"engram": {
"command": "engram",
"args": ["serve"]
}language mix
live snapshotStart local. Scale into teams.
Keep the first step light, then unlock shared graph workflows, cloud sync, and team controls when Engram becomes part of the way you ship.
Free
$0solo
Pro
$10/month
Team
$30/dev/mo
Give your agents memory, context, and vision.
Engram is the layer between your repo and every coding agent that touches it. 16 tools, persistent memory, blast-radius analysis, and a VS Code extension — stop paying the same warm-up penalty on every task.