FAQ

Who

Who is ContextBrain for?

Three buyers, one promise. Individual developers and engineering teams who want AI that is fluent in their codebase; agencies and consultancies that onboard to unfamiliar client code fast while keeping each client's IP isolated; and enterprises that need RBAC, audit, compliance, and on-prem deployment. See who it's for.

Do I need to replace my IDE or my AI model to use it?

No. ContextBrain is the layer underneath the tools you already use — bring your editor, your model, and your agent. It is model- and agent-agnostic by design.

Does it work with my IDE?

Yes — any IDE or agent that speaks the Model Context Protocol (MCP) connects to ContextBrain, including Claude Code and Cursor. There is also a REST API and an embeddable chat widget. See chat with your codebase.

Why

What problem does ContextBrain actually solve?

AI coding tools have never seen your codebase, so they guess — and the guessing shows up as hallucination, wrong-file edits, stale context, and amnesia. ContextBrain assembles the right context per task so the model stops guessing. Read what ContextBrain is.

Isn't this just a model problem that better models will fix?

No. The frontier models are extraordinary; these are contextfailures, and context is infrastructure you have to build. Google's DORA 2025 research independently states the thesis: AI is an amplifier, and the return comes from the underlying system and AI-accessible context — not the tool itself.

How is this different from Copilot or Cursor?

Copilot and Cursor are IDE assistants — great at in-editor flow for one developer in one repo. ContextBrain is the context-and-memory platform underneath every agent: multi-repo, governed, auditable, with task-scoped context packs and an SDLC loop. It completes those tools rather than replacing them. See Alternatives.

How

How does it understand my code?

It indexes every repository into a tree-sitter code graph plus hybrid retrieval — semantic embeddings, BM25 keyword, and entity matches fused with Reciprocal Rank Fusion. See the knowledge graph and the end-to-end pipeline.

How fresh is the context?

Indexing is incremental on every commit (webhook-driven), with on-demand re-index and a per-query cache that auto-invalidates when a repo re-indexes. Every context pack is scored for freshness, so you can see whether context is two hours or two months old before acting.

What is a context pack?

The unit the AI actually consumes: a task-scoped bundle of code chunks + memory + tickets + docs + your coding guidelines + similar prior work, scored for confidence and freshness, pruned to the model's token budget, and hash-signed so you can prove what the AI saw. See context packs.

Which AI models does it work with?

You bring your own provider keys — OpenAI, Anthropic, Voyage, Cohere — and select models per slot. Agents support Claude, Codex, Gemini, and Aider. When a better model ships, you switch a dropdown, not a platform.

Does it support Laravel / PHP?

Yes. ContextBrain is positioned as the first context platform that truly understands Laravel/PHP codebases, alongside the rest of its language support. See Laravel & PHP.

Cost

What does it cost?

The platform runs on three plan tiers — Starter, Growth, and Enterprise — and you pay your own AI bill with no markup on top. See pricing and Cost.

Do I pay a markup on AI tokens?

No. ContextBrain does not resell tokens. You connect your own provider keys and every call bills to your account at your provider's rate — there is no per-token surcharge between you and the model.

Can I self-host?

Yes. ContextBrain is Postgres-native and runs via docker compose, with a self-hosted daemon mode for agent execution, a local reranker, and a local-embeddings option — which can drive vendor spend toward zero. See Cost.

Can I see what AI is costing us?

Yes — an org spend tile on the dashboard, a per-project cost page with cost-by-agent and cost-by-model breakdowns, and a billing rollup by project and model. AI spend becomes a managed line item, attributable per client and per project.

Risks

Does my code leave my servers?

Only if you let it. In self-hosted daemon mode, agent execution, reranking, and embeddings run inside your perimeter, so the codebase never has to leave. Per the research, on-prem / air-gapped deployment is the primary buying driver in regulated sectors — and a segment cloud-only tools like GitHub Copilot and Amazon Q Developer structurally cannot serve. See Risks & trust.

How do you keep one client's code from leaking into another's?

Client-safe context redaction strips proprietary code and secrets before context reaches an external AI API, and every client's context is isolated. IDOR-safe cross-tenant checks guard every project, agent, and squad reference. See security.

Is it secure enough for a security review?

That is the design target: AES-256-GCM at rest for every secret, RBAC enforced via canPerform, a 90-day audit log with secret scrubbing, prompt-injection delimiters, and fail-secure webhooks. See Risks & trust.

What about GDPR and data deletion?

ContextBrain has a built-in right-to-be-forgotten flow: request → 30-day grace → purge, anonymizing the user row in place while preserving audit foreign keys. Audit retention is 90 days.

Does it really stop hallucinations?

It reduces them sharply, but does not eliminate them — and we say so plainly. Grounding a model in the right files, schema, and decisions collapses most guessing, and every answer cites its sources so you can check the claim against the citation. AI still needs human review. See the honest limitations in Risks & trust.

Alternatives

Do I have to stop using my current AI tools?

No. Many teams run ContextBrain alongside their IDE assistant or autonomous agent — it makes those tools fluent in your codebase rather than replacing them. See Alternatives.

How is this different from a code-search tool like Cody or Greptile?

Search answers “where is this?” ContextBrain assembles a task-scoped, scored, token-budgeted context pack and adds governed memory, agents, and an SDLC loop on top. Search is one ingredient; ContextBrain is the meal. See Alternatives.

Isn't MCP the thing that already solves this?

MCP is the open standard for feeding context to agents — the pipe. ContextBrain is the brain that decides what is worth sending through it. They are complementary: ContextBrain speaks MCP natively. See chat with your codebase.