AIDEAID Knowledge · v1.2 · as of 2026-07-15 · a snapshot; this market moves fast
Our homepage makes a big promise: your knowledge stays with you, and the AI on top is a swappable module. Everyone writes sentences like that these days. So we put ours to the test, platform by platform: what would it actually mean to move an AI system, the kind AIDEAID builds, to a different provider? Seven switching options plus our current baseline and one market category for context, each researched individually and cross-checked in a second, independent review pass.
Tip: tap a card for details (capabilities, data protection, cost, cross-check, sources). The sort tabs reorder the comparison by each metric.
1Key findings
The lock-in is small, and it is measurable.Only roughly 10–20% of an AIDEAID system depends on the AI provider. Over 100 automation building blocks, all your knowledge in open plain-text formats, the database, and the scheduling keep running on any platform.
The cheapest switch is not a migration.The AI model can be swapped underneath the running system: about 3 working days via compatible cloud endpoints. Converting to the fully local track is a delivery model of its own, realistically ~20 working days. A complete platform switch costs 14 to 40, depending on the target.
Open standards work in your favor.The three load-bearing building blocks have been open standards at the Linux Foundation since late 2025: the work instructions (skills), the central context file (AGENTS.md), and the connection to the knowledge archive (MCP). OpenAI, Google, Mistral, and xAI already support them. Portability grows every month.
And today?We build on Claude because its performance and tooling maturity currently set the benchmark. OpenAI would be the most mature plan B, Mistral the EU alternative, the local track the option for maximum confidentiality. That freedom of choice is exactly the point.
2How much really depends on the AI provider?
An inventory of a complete AIDEAID system, roughly weighted by effort. Expand to see what sits in which bucket.
~70% runs anywhere
~20%
Portable, moves 1:1Portable with reworkNeeds to be rebuilt
Portable: the substance simply moves over
The automation building blocks: over 100 scripts that collect data, generate reports, sort mail. They talk to the AI directly through the interface; the model behind it is configuration.
All your knowledge: notes, processes, context, as Markdown plain text. Readable with any editor, usable by any AI model.
The business data: SQLite, an open format with 25 years of history.
The scheduling: built-in macOS tooling, no platform magic.
Web applications (portals, dashboards): run on your own EU server, completely provider-independent.
The connection to the knowledge archive: MCP, an open standard under the Linux Foundation. The connector stays, no matter which model is attached.
With rework: the texts stay, the wiring is new
Work instructions (skills) and shortcut commands: the texts are finished material, the wiring differs per platform. Thanks to the shared open skills format, increasingly mechanical work.
The central context file, the system’s “world knowledge”: content portable, loading mechanics platform-specific.
Startup automation and safety guardrails: the scripts run anywhere, the mounting points do not.
The long-term memory: the memories are simple files, the automatic recall is a platform feature.
Rebuild required: the platform specialties
The platform’s own control mechanics: event hooks, permission system, command dispatch.
The orchestration of parallel AI agents (deep research runs, multi-pass reviews).
Running automations on a subscription: a flat rate instead of an open-ended usage bill through the technical interface (API). Surprisingly few providers do this cleanly.
Convenience features like remote access to the running system from your phone.
3The platform comparison
Four metrics per platform. Score bars (blue): longer = better. Effort bars (warm): shorter = better. Baseline anchor: ~6 setup days for a new installation on our current standard platform, with a proven approach. All day figures are estimates, not offers. The scores are our editorial assessment, not measurements.
Sort by:
Model performance (0–10)Agent tooling capabilities (0–10)New setup (working days)Migration from the running system (working days)
4What does this mean for you?
We build on Claude, because performance is the deciding factor. Independent coding rankings (including LMArena, as of spring 2026) currently place the Anthropic models on top. In our assessment, the alternatives we examined cost quality, working days, or both.
The ability to switch is built in. Your knowledge lives with you in open formats, and the AI model can be swapped underneath the running system. The foundation holds even as the AI world keeps turning.
For EU sovereignty requirements, Mistral stands ready as a serious European second platform, at a fraction of the cost, one performance class below.
For maximum confidentiality (law firm, medical practice) there is the local track: your own hardware, data stays physically on premises. A product of its own with its own set of expectations.
Methodology: Researched on July 13, 2026 from current primary and secondary sources (provider documentation, independent benchmarks, trade reporting). Each platform was cross-checked in a second, independent review pass; corrections are incorporated and transparently noted on the cards under “Cross-check”. Some benchmark figures are vendor-reported and labeled as such. Working days are estimates relative to our proven setup approach. The scores (0–10) are our editorial assessment based on the cited sources, not measurements; the baseline platform is set to 10 as the reference point. This page is a snapshot: the AI market moves week by week.
Questions about any of these assessments, or newer findings? Write to us: axel@aideaid.ai
