01Efficiency
Spend fewer tokens on context
Every agent starts with the current code structure and the decisions behind it.
Better agent output with less context waste
Get more useful workfrom every coding agent.
Memtrace gives agents the code map and decisions they need before editing. Your team spends fewer tokens on context and catches risk before review.
✓Public benchmark: 3.9× more correct results per response token than the runner-up.
Product walkthroughexample repository
01
See how the codebase fits together.
Follow each function to the code that calls it and the systems it depends on.
Works with the coding tools your team already uses
CodexClaude CodeCursorGemini CLIWindsurfVS Code+ more
What your team gets
Get more useful work from every agent.
Spend fewer tokens getting there.
Memtrace cuts the follow-up searches that burn tokens and surfaces change risk before review.
84.7%fewer file-content tokens
Typical session compared with raw reads
202response tokens per correct result
Public 1,000-query exact-symbol benchmark
3.9×more correct results per token
See methods and raw resultsCompared with the next benchmarked tool
02Risk
Catch risk before review
Trace affected code and services before a risky edit becomes review work.
03Continuity
Stop losing key decisions
New engineers and agents can see why the code works this way.
04Throughput
Prevent work collisions
Spot overlap before merge. Bring people in only when judgment is needed.
One product for every coding agent
Map the codebase once.
Reuse the context across your agents.
Memtrace connects code paths and decisions so agents can trace impact and spot overlap without searching the repository from scratch.
Know the codebase
See how the whole system fits together.
Memtrace builds a live map of the codebase, from functions and callers to APIs and execution flow. Engineers and agents can trace impact without piecing it together from search results.
- ✓Blast radius before edits
- ✓Cross-repo API paths
- ✓Callers and execution flow
What changes with authorizePayment?Caller path and covering test found
See how Memtrace maps codeMemtrace found the full request path and existing coverage.
How it fits your workflow
Connect one repo.
Give every agent the current code map.
Memtrace connects to coding tools through MCP. It keeps a current code map available to each agent, so they can check impact and team decisions before editing.
CONNECT
Connect one repository.
Sign in, receive your license, then connect your coding agent through MCP.
STAY CURRENT
The code graph updates with your repository.
Memtrace maps functions and dependencies, then keeps history and recorded decisions attached as the code changes.
CHECK FIRST
Agents check before changing code.
Before editing, agents can trace impact and recall why the code exists. They also see who else is working on the same code.
Questions your team can answer
Ask about a change.
See the code behind the answer.
Each example uses the same live code graph and recorded decisions.
ASK MEMTRACE
IMPACT ANALYSISExample result
›What breaks if we change auth?
12 callers across 2 services
The session refresh path reaches the API gateway and billing worker. Three tests cover the affected boundary.
auth/session.tsapi/gateway.tsbilling/worker.rs
✓Answer checked against the code graph and recorded history
Deployment and control
You choose where Memtrace runs.
Source analysis runs locally by default. Teams can run shared memory in their own infrastructure while keeping decisions and parallel work connected to the codebase.
See team deployment optionsHow Memtrace fits your environmentLocal analysis to team coordination
Public benchmark results
Find the right code.
Keep up with every edit.
On 1,000 exact-symbol queries, Memtrace puts the target file first 96.6% of the time. Separate runtime tests measure lookup, freshness, indexing, and memory.
View methods and raw resultsRetrieval quality
1,000 scored queriesThe right file, ranked first.
Top-1 accuracy96.6%
Target file ranked first.
Precision@10
96.7%Penalizes extra paths around a correct hit.
Recall@10
99.7%Target file found in the first 10.
MRR
0.979Scores an earlier first correct result higher.
Runtime
4 public testsBuilt for the coding loop.
0.19 ms
average graph lookup
Django · 50K-node graph
p95 update-to-query time
50-edit incremental fixture
repository to queryable graph
Django · 3.3K files
peak process memory
isolated Django process
Agent benchmark · ContextBench
500 tasks · 8 languagesIt's not the model. It's the memory.
Your agent doesn't fail because it isn't smart enough. It fails because it's guessing which 200 lines of a 400,000-line repo matter — so you upgrade the model and pay more to keep guessing.
gpt-5-mini + MemtracebeatsClaude Sonnet 4.5
2.5×
cheaper per task$0.30 vs $0.76
- AgentlessGPT-50.376
- Memtracegpt-5-mini0.3465
- mini-SWE-agentClaude Sonnet 4.50.344
Reaches 92% of Agentless + GPT-5's context quality running a small model — because Memtrace answers with lexical, semantic and structural retrieval in one query, instead of a keyword search bolted to a vector database.
Open any result to see the repository, test method, and raw output.
Start with one repository
Sign in once.
Start with your next change.
Your account and Community license are created immediately. Then run Memtrace at your repository root and connect the coding agent you already use.
PATH 01 · INDIVIDUAL
Create your account before you install.
Sign in with Google or GitHub. Memtrace creates your Community license, then guides you through the first repository.
your-repo / terminal
01 Sign in with Google or GitHub
02 Community license created
03 npx memtrace@latest start
Repository indexed · agent connectedShare codebase context across the team.
Give engineers and agents the same graph context. Run shared memory in infrastructure your team controls.
- ✓Self-hosted MemDB
- ✓Fleet coordination
- ✓Decision governance
- ✓Cross-repo topology
Before rollout
What teams need to know.
The docs cover authentication and exact CLI behavior. Self-hosted deployment boundaries are documented there too.
Read the documentation↗No. Memtrace connects to your existing coding agents through MCP. It lets them query the code graph and reuse project decisions before and during a change.
MemtraceThe context should outlive the chat