Cost

The model: bring your own keys

ContextBrain is model- and agent-agnostic by design. It does not resell tokens. You connect your own provider keys — OpenAI, Anthropic, Voyage, Cohere — per organization, encrypted at rest with AES-256-GCM, and every embedding, chat, and agent call bills to your account at yourprovider's rate. There is no per-token surcharge sitting between you and the model.

Two consequences fall out of that. First, there is no lock-in at the model layer: when a better or cheaper model ships next month, you switch a dropdown, not a platform. Second, your AI spend is exactly what your provider charges — ContextBrain's job is to make each call do more with fewer tokens, which is a cost lever rather than a cost center.

Self-host to zero vendor spend

ContextBrain is Postgres-native. The full relational core, the code graph, the memory tiers, and the cost telemetry all live in a standard Postgres database — no proprietary datastore you have to license. You can stand the whole stack up with docker compose on hardware you already own.

In self-hosted daemon mode, agent execution, a local reranker, and a local-embeddings option all run inside your perimeter. For teams that already run their own inference (a local model server, an on-prem GPU box), this means the platform cost can be effectively zero vendor spend — you pay for the hardware you were going to run anyway, and your code never has to leave the building. On-prem deployment is the primary buying driver in regulated sectors; see Risks & trust for the security posture behind it.

Token economics

The pricing of AI tooling is shifting from flat per-seat fees toward usage- and token-metered billing. Per GitHub's announcement, all GitHub Copilot plans move to usage-based billing on June 1, 2026, replacing premium-request units with token-consumption-based credits (base seats unchanged: Pro $10/mo, Business $19/user/mo, Enterprise $39/user/mo). As billing goes token-metered industry-wide, context efficiency becomes cost control.

This is where ContextBrain's architecture pays for itself:

• Token-budget pruning— context packs are pruned to fit the target model's window. You send the model the right context, not the whole repository, so you are not paying to stuff irrelevant code into a prompt.
• Pack caching — packs are fingerprinted and cached by (project, task, depth, agent_type). Two developers asking similar questions share a cached pack instead of regenerating (and re-paying for) the same assembly.
• Grounded answers cut rework— the expensive failure mode is the “almost-right” answer you have to find and fix. Per the 2025 Stack Overflow Developer Survey (49,000+ respondents), 66% of developers now spend more time fixing almost-right AI code. Better context reduces that re-spend.

The cost story and the quality story are the same story: a model that works from the right context guesses less, retries less, and burns fewer tokens doing it.

Plans

ContextBrain organizations run on one of three plan tiers — Starter, Growth, and Enterprise. Starter is the entry point for a single team getting fluent on one codebase; Growth suits agencies and multi-project orgs that need multi-repo, multi-client isolation and cost visibility across projects; Enterprise adds self-hosted / air-gapped deployment, governance, and the controls a security review expects.

Across every tier the model bill stays yours — the plan covers the platform, not your tokens. For the current tier breakdown and what each includes, see the pricing page.

Cost visibility

Because the model spend is real money on your account, ContextBrain makes it a managed, visible line item rather than a surprise. Cost telemetry is recorded per session — input, output, and cache-read tokens — and surfaced at three levels:

• Org spend tile on the dashboard — month-to-date spend with a last-month delta, so leadership sees the trend at a glance.
• Per-project cost page — metric cards, a daily line chart, cost-by-agent and cost-by-model bars, and a top-sessions table, so you can see exactly which work and which model drove spend.
• Billing rollup — a live-rollup card plus stacked bar charts by project and by model, the FinOps-friendly view for an engineering leader.

Together these turn AI spend from an opaque vendor invoice into something you can attribute per client, per project, and per model — which is exactly what an agency billing clients, or an enterprise running a budget, needs.

Where to go next

The honest counterpart to the cost story is the trust story — read Risks & trust for the security posture and candid limitations, compare the field on Alternatives, or skim the quick answers in the FAQ.