If you picked an AI coding tool specifically because it could run against a local model — no code leaving your machine, no vendor in the loop — the list of options just got shorter twice in three months, and neither correction seems to have made the rounds yet.
Tabnine was the default answer here for years: self-hosted, enterprise-friendly, model-agnostic. Tricentis acquired it on 30 July 2026. tabnine.com now redirects straight to Tricentis's site, there's no public pricing page anymore, and the only people it still serves are existing customers — new signups are gone. We checked this directly against Tricentis's own announcement rather than trusting secondhand coverage, because "acquired" gets thrown around loosely and we wanted to know specifically whether new customers could still buy in. They can't.
Cody is the other half of this. Sourcegraph made it enterprise-only back on 23 July 2025 — if you missed that one, you're not alone, it's well over a year old now and still surprises people. Minimum contract is around $16K/year per Sourcegraph's own pricing page. It's still the only tool we track that sells an actual self-hosted deployment for a whole team, so if that's a hard requirement and you have the budget, it remains the answer. For an individual developer, it simply isn't purchasable anymore.
That leaves two tools that were never venture-funded SaaS products trying to monetize an enterprise upsell, which turns out to matter: Cline and Aider. Both are open source, both point at local inference through Ollama or LM Studio out of the box, and both will happily run against a hosted API instead if you'd rather trade privacy for a stronger model — the point is that the choice is yours, not baked into the product. Cline runs in VS Code with Plan/Act modes and MCP support; Aider is terminal-native with tight git integration if you don't want an editor in the loop at all. We rate Cline as the stronger overall pick for this use case, Aider as the better fit if you live in a terminal.
Worth naming as a fourth option even though it's not strictly "privacy-focused" by itself: Zed, a Rust-native open-source editor that's agent-agnostic — it doesn't force you through one vendor's AI layer, so you can wire up whatever inference path you want, local or not. Its own Personal tier is free.
The honest summary: the self-hosted product category is contracting (two fewer options than six months ago), while the self-hosted capability — bring your own local model — is fine, because it never depended on a vendor selling you a specific deployment story in the first place. If you're re-evaluating because your old pick just got acquired or locked behind an enterprise contract, that's the distinction to make before you pick a replacement.
Full comparison table (15 tracked dimensions across all five tools, prices re-checked against each vendor's own page) is here: https://devtoolsreview.com/best-for/best-ai-coding-tools-privacy-focused/
Disclosure: we earn nothing from any link in this piece. We're not enrolled in an affiliate programme for any of these five tools — most of them don't run one. Facts checked directly against vendor pricing pages and Tricentis's own acquisition announcement, current as of late October 2026.
Top comments (0)