Access policy
Sentinel matches repo-relative paths and either denies the read or returns it redacted, with the reason you wrote in the rule. Denied paths drop out of search results too, so an agent cannot reach guarded code through retrieval either.
The SourceVault Gateway
Governed AI access to private code.
SourceVault is the gateway between coding agents and the code they are allowed to read: an access policy, secret redaction, a tamper-evident audit chain, signed provenance, and named agent identities, enforced once on every surface. Underneath the gate, agents get what makes them effective: grounded, file-and-line-cited answers over your code and its entire git history, on hardware you already own. Nothing is sent to a third party.
What buyers get
- SourceVault Gateway is the governing layer: an access policy over what AI can read, and secrets redacted from every answer. It comes with every license, and the 7-day trial runs it on one repository.
- Every access lands in a tamper-evident audit log; AI-authored changes carry signed provenance.
- An operator console shows what agents accessed and what was stopped, and verifies the audit chain in one click.
- A dashboard with cited answers, a full source viewer, watches, and pinned answers exported as a runbook.
- Git-history answers and module overviews, with retrieval tuned for agents. A cloud indexer sees none of it.
- SourceVault indexes GitHub, GitLab, and Bitbucket repositories, as well as any folder on disk.
- An MCP server for any MCP client, plus the Hermes plugin; every surface sits behind the same gate.
- One-command install on macOS, Linux, or Windows (WSL2). The engine is free forever on one repository, with a 7-day trial of the Gateway on it.
The problem
Hosted AI tools break down when the code is private.
Most code assistants upload your repository and answer from a lossy slice held in a cloud context window. On a large or sensitive repo, that creates privacy and compliance risk, and the answers get weak.
And even when the tooling is private, a security owner still has to answer for it: which repos and files can agents touch, and what exactly did the AI read last quarter? SourceVault is built for teams that need to permit AI on private code. It answers both questions on infrastructure you own: policy decides every read, and the log shows what the AI actually saw.
The Gateway
Permit it, bound it, prove what it read.
Before a coding agent gets near a private repository, someone has to answer whether you can permit it, bound it by policy, and prove afterwards exactly what the AI read. The SourceVault Gateway answers all three, and it runs in the engine's data path rather than beside it: the dashboard, the MCP server, the signed machine API, and the Hermes plugin all call the same tool layer, so one policy evaluation governs every surface and every call lands in the same audit chain.
Secret redaction
Every file read, search result, and answer passes a gitleaks-style scan first. A key in a config file comes back as a redaction marker before the content reaches the agent, and therefore before a cloud-backed client could forward it to a model.
Audit chain
Each ask, search, file read, and policy decision appends one hash-chained record, anchored by a signed head, so truncation or deletion is detectable. The operator console reads it back and verifies the chain in one click; the chain exports to your SIEM as JSONL, CEF, or syslog, and a period report states what AI read, wrote, and was stopped from.
Named agents, signed changes
Agents get scoped tokens bound to the surfaces and repositories they may reach, so the chain says who asked, not just what was read. AI-authored, security-relevant commits carry signed provenance you can verify offline.
Underneath the gate: the engine
Grounded answers from exact reads of a local index.
What the gate hands a permitted agent is what makes it effective: grounded, file-and-line-cited answers over your code and its git history. The engine implements textbook retrieval patterns, built directly on a local SQLite vector index and Ollama, with an optional Qdrant engine that SourceVault runs for you when full git history takes a repository to serious scale. The stack stays small enough to audit line by line.
Hybrid retrieval
Each query combines semantic search and exact keyword matching, then fuses the rankings so the strongest results rise to the top.
Code-aware chunking
Files are split along function and class boundaries, so every result maps to readable code with exact line ranges.
Context-aware embeddings
Each chunk is embedded with its path and symbol names, so the vector store knows where each piece of code lives in the repository as well as what it says.
Grounded answers
Ask mode checks whether it has enough context, retrieves again if needed, and answers only from source-backed snippets instead of guessing.
Cost model
Retrieval is cheaper than re-reading the repo.
An AI agent will re-read a whole file to find one function, then drag that stale context forward turn after turn. The waste compounds with every question a developer asks, and metered plans bill all of it. SourceVault answers from a bounded budget of cited file and line ranges instead. Repeated questions return from cache without a model call, and local models have no meter at all.
Metered feesNone. Starter and Pro are flat one-time payments; Team is a flat annual seat count. Nothing is billed per token
Per-token billingNone for local answers; repeated questions return from cache with zero model calls
Context per answerA bounded budget (about 6k tokens by default) of cited file and line ranges
Source code egressNone. Nothing is uploaded, logged, or retained by a third party
Beyond code search
Governed access first, then everything the engine adds.
Governance first, then an engine that tracks how your code connects and re-checks its answers as the code changes. All of it is driven from a browser dashboard the whole team can use. A few highlights:
SourceVault GatewayEvery tier
Governed AI access to your code, in every license. The Sentinel policy gates which files can be read at all, a DLP pass redacts secrets from every answer before delivery, each decision lands in a tamper-evident hash-chained audit log, and AI-authored security-relevant commits get signed provenance attestations. Standard on every license; the 7-day trial runs it on one repository, and the free tier keeps the engine on that repository afterwards.
100% local by design
Ollama generates the embeddings and a local SQLite file stores the vectors, both on your own machine, with no vector database service to run. The model that writes the answer runs there too.
Git history answers
Commit messages are indexed alongside code, so questions like "why was this changed?" and "when did this break?" are answered from the actual commits. Cloud indexers can't do this, since they only ever see a snapshot of the tree.
Multi-repo AskPro
Ask one question across every indexed repository at once, with each citation tagged by repo. "How do the frontend and backend handle this?" comes back as a single answer. Standard from Pro up.
Watches
Pin a question as a standing check. After every reindex it runs again and flags you if the cited answer has drifted. Useful for questions like "did the auth flow change this sprint?"
Works with your AI tools
An MCP server exposes the same engine to Claude Code, OpenClaw, and any other MCP client, which get bounded, cited context instead of re-reading files. The server itself is local-only. Pair it with a local-model client and the whole loop runs offline; a cloud-backed client sends what it retrieves to its own vendor, by your choice.
Pricing
Free on one repository. License the Gateway when you need it.
The engine is free forever on one repository. Every license carries the SourceVault Gateway, the same governed path the 7-day trial runs; what changes between tiers is the number of sources you index, where a source is a repository or a local folder, and from Pro up, Multi-repo Ask. Starter and Pro are flat one-time payments and Team is a flat annual seat count: no per-token metering. Loose files you add individually share a single source slot.
Start free
One command installs the engine (search, read, Ask), free forever on one repository, with the SourceVault Gateway running on that repository for 7 days. No account, no card, nothing expires. If it can't answer questions about your code with file-and-line citations, don't license it.
Changed your mind? One question before you go.
Which half of SourceVault brought you here? Your answer decides what we build next, and it is the only thing we record.
Thank you. That is exactly what we needed to know.
Starter
$390 one-time
For one engineer putting coding agents on their own private repos, with the Gateway in front of them.
- Up to 3 indexed sources (repos or local folders)
- One machine
- Dashboard, Hermes plugin, MCP server
- Git-history answers and module overviews
- Webhook-driven reindex on every push; watches that re-check themselves
Pro
$4,500 one-time
For a small team running AI across several private repos, and asking questions that span them.
- Multi-repo Ask included
- Up to 15 indexed sources (repos or local folders)
- Everything in Starter
Team
From $15,000/year
The Gateway rolled out across a team: every repository, every agent, one policy, with a guided setup.
- Annual per-seat term, unlimited sources
- Team support included (the Priority Support service level)
- Security-focused file exclusions and a team onboarding runbook
- Everything in Pro
Enterprise
Custom
For regulated or air-gapped environments that need the compliance story in writing.
- Air-gapped install: offline bundle and models
- Unlimited sources and seats
- The compliance pack and offline delivery, in writing
- Priority Support SLA
In every tier
SourceVault Gateway
An access policy over what AI can read, secrets redacted from every answer, a tamper-evident audit chain, signed AI-change provenance, and named agent identities. The same governed path the 7-day trial runs.
Operator console
What agents accessed, what was stopped, and chain verification in one click.
Updates included
Starter and Pro include 12 months of updates; the version you install is yours forever, and renewing later costs a fraction of list price. Team includes updates for the whole term.
A key, not an account
The license key arrives by email, activates instantly, and verifies offline. No login, no phone-home. The free tier needs no key at all.
Add-ons
Priority Support Subscription
The one optional subscription: a support SLA plus hands-on help, model and resource tuning for your hardware, and reindex strategy when embedding models change. Flat $195/mo for Starter and Pro customers; included in Team. Requires an active SourceVault license (buy with the same email). Cancel anytime.
Updates renewal Existing customers
Your license and installed version work forever. When your 12-month updates window ends, renew to keep receiving new releases: Starter $160, Pro $1,850, one-time. The fresh key arrives by email and replaces the old one in the dashboard's License settings.
Team is quoted per seat before work starts, and you get a straight recommendation even if the right answer is the smaller package. Every purchase carries a 14-day money-back guarantee (Team: on the first term). Licenses and Team terms never auto-charge; the optional support subscription recurs monthly until you cancel it, and a lapsed Team term falls back to the free tier. The details are in the billing & refund policy. Starter customers can upgrade to Pro any time for the difference in list prices ($4,110 today) on the upgrade page (Settings → Features links to it), or email support.
Deployment
One command on every platform.
The models, the index, and the dashboard all run on machines you control. One command installs everything on macOS, Linux, or Windows (WSL2), or use the shared Docker Compose deploy where teammates on any OS connect through the browser. Every install is free on one repository, with the 7-day Gateway trial included.
Questions?
Talk to a human before you buy.
Team and Enterprise rollouts, air-gapped installs, checkout trouble, or anything the trial can't settle: email us and you'll get a straight recommendation, usually the same day. For install help, Discord is fastest.