04 / AI SYSTEMS COMPARED
RiotSpark vs ChatGPT, Gemini, Copilot, Claude and Letta
Where does a self-owned AI thinking partner fit?
RiotSpark puts persistent identity, governed memory and human authority at the centre of its architecture. This comparison explains that focus alongside established assistants and stateful agent platforms, with a clear boundary between working features and future plans.
By Raziel Delacroix · RiotSpark status updated 2 October 2026 · Competitor sources reviewed 1 October 2026 · Architecture comparison, not a performance benchmark
Different products. A useful comparison.
ChatGPT, Gemini, Copilot and Claude are assistant products. Letta is a platform for stateful agents. RiotSpark is an independently developed AI core that uses a language model as its reasoning engine. Comparing them is useful when the question is how memory, identity and control work—not which name wins an intelligence score.
The established assistants already offer memory and personalisation. Letta also supports persistent identity, local deployment and model choice. RiotSpark’s focus is the combination of a protected Core, reviewable knowledge, approval rules and recorded reasons for remembering.
Scope: the named assistant products and documented Letta capabilities, not every provider API, enterprise deployment or third-party extension. Features vary by plan, region, account and configuration. Descriptions below link to official documentation; absence of a feature from this page does not establish that a product lacks it.
AI memory, identity and ownership at a glance
Read each product in its own context. On smaller screens, scroll the table horizontally.
RiotSpark and five related AI products — documented capabilities and development status| System / purpose | Memory and continuity | Ownership and scope | Evidence |
|---|
| RiotSparkIndependent AI core in development | RiotSpark 2.0 separates identity, Agnes Personality 1.0, durable conversations and governed memory. Memory informs as reference evidence and cannot silently redefine behaviour. | Provider-neutral adapters are implemented. Ollama with Qwen3 1.7B is the primary local path; GPT is used as a reference path. | Working interface · Project milestones |
|---|
| ChatGPTHosted AI assistant | Memory can use context from chats, files and connected apps. Users can review and correct a memory summary and inspect which sources personalised a response. | Users can delete contributing sources or disable memory. Exported conversations can be uploaded elsewhere as reference, but this does not recreate the original account, history or independent identity. | Memory · Transfer |
|---|
| GeminiGoogle’s AI assistant | Eligible personal accounts can use past-chat memory, response instructions and selectable Gemini models. | Google documents memory and chat-history import from other AI platforms, although that import is currently unavailable in the UK, EEA and Switzerland. | Memory · Models · Import |
|---|
| Microsoft CopilotMicrosoft’s consumer AI assistant | The current consumer Copilot app can save personal details and preferences as memories. Signed-in users can review and delete saved memories, delete all memories or disable further memory storage. | This comparison covers consumer Copilot, not Microsoft 365 Copilot or GitHub Copilot. | Microsoft documentation |
|---|
| ClaudeAnthropic’s AI assistant | Claude can maintain topic-based cross-chat and project-specific memory, with controls to pause or reset it. | Memory import and text-based export are documented, although import remains experimental and may not retain every supplied detail. | Memory · Transfer |
|---|
| LettaStateful agent platform | Persistent, versioned memory through MemFS, including identity files and on-demand reference material. AgentFile can transfer persistent agent state, memory and behaviour between Letta servers. | Supports local or self-hosted agents and multiple model providers. Local-only MemFS repositories remain the operator’s backup responsibility. | AgentFile · Memory · Hosting · Models |
|---|
What RiotSpark does today—and what comes next
ACTIVE LINE · RIOTSPARK 2.0A personality-first governed system
RiotSpark 2.0 includes an identity contract, Agnes Personality 1.0, context classification, behaviour compilation, provider-neutral model adapters, persistent conversations, governed memory, recovery and safe mode.
P2.5 added tier-aware retrieval and storage visibility. P2.6 added review, governed editing, correction, supersession and preserved audit history. Milestones A–L established and froze the personality-first foundation. M selectively adapted useful V1 capabilities; N and O added portability, recovery and stability; P added durable conversation threads.
The latest reviewed acceptance record confirms R2.0-P-E after automated and live validation, with 364 tests passing. Agnes Personality 1.0 is frozen.
View completed milestones →IN VALIDATION · R2.0-P-F — QSeal persistence and prove the candidate
The persistence seal is tracked and RiotSpark 2.0 release-candidate materials are being prepared. Validation covers persistence, recovery, documentation and the assembled candidate boundary.
P-F and Q are not presented as accepted until a formal acceptance record exists. RiotSpark V1 remains preserved at P2.8.7, with P2.9 paused.
Follow the current roadmap →LONGER-TERM DIRECTIONContinuity across richer interfaces
The aim is to preserve RiotSpark’s identity as reasoning engines and interfaces evolve. Voice, visual presence, embodied interfaces and robotics remain future directions.
No claim of universal model compatibility, autonomous company operation or superiority over other assistants is made here. Model changes still need compatibility and behaviour checks.
Read the governing principles →
Memory controls, model choice, portability and self-hosting
These terms can sound interchangeable, but they describe different kinds of control. Memory controls determine what a system remembers and what a person can review, correct or delete. Model choice determines which reasoning engine can be used. Portability determines what can move between accounts, servers or runtimes. Self-hosting determines who operates the software and carries responsibility for updates, security and backups.
ChatGPT, Gemini, Copilot and Claude provide increasingly capable memory controls inside hosted assistant products. Their documented transfer features primarily move conversation or memory material. Letta goes further architecturally through self-hosting, multiple model providers, MemFS and AgentFile portability. RiotSpark 2.0’s implemented distinction is a personality-first authority stack outside the model: identity defines the system, Agnes governs behaviour, and memory remains reference evidence. Provider adapters are separated from those authorities.
How RiotSpark compares with each system
RiotSpark vs ChatGPT
ChatGPT can use context from chats, files and connected apps, while letting users review or correct a memory summary and inspect personalisation sources. Exported conversations can be uploaded elsewhere as reference, but that does not recreate the original account or an independent identity. RiotSpark concentrates on keeping identity and approval in an owner-operated Core outside the model. [1] [2]
RiotSpark vs Gemini
Eligible Gemini accounts can use past-chat memory, response instructions and selectable models. Google also documents import from other AI platforms, but that feature is unavailable in the UK, EEA and Switzerland. RiotSpark’s current implementation focuses on a local runtime and explicitly governed knowledge capture. [3] [4] [5]
RiotSpark vs Microsoft Copilot
The current consumer Copilot app can save personal details and preferences as memories. Signed-in users can review or delete saved memories, delete all memories or disable further memory storage. RiotSpark instead makes its protected Core and knowledge approval process part of its own runtime. This comparison excludes Microsoft 365 Copilot and GitHub Copilot. [6]
RiotSpark vs Claude
Claude documents topic-based cross-chat and project-specific memory, with controls to pause or reset it. Import and text-based export are also documented, although import remains experimental and may omit information. RiotSpark’s aim is to preserve the governing system itself across engine changes, rather than only move remembered text between services. [7] [8]
RiotSpark vs Letta
Letta is the closest architectural comparison here. Its MemFS stores versioned memory, while AgentFile can import and export persistent agent state, memory and behaviour between Letta servers. It also supports local or self-hosted agents and multiple model providers. Local-only MemFS repositories remain the operator’s backup responsibility. These are real areas of overlap, not exclusive RiotSpark features. [9] [10] [11] [12]
RiotSpark’s emphasis is its particular governance policy: protected identity, owner approval, knowledge classification and a mandatory why-layer. Whether that policy serves a use case better requires practical evaluation. Letta can be extended; this page does not claim it cannot implement similar controls.
Why follow RiotSpark?
If you care about a persistent AI companion whose identity, memory rules and development direction remain under human control, RiotSpark is a project to watch. Its public value today is a working local foundation and a visible development path. It is not presented as a finished replacement for the established products above.
No comparative accuracy, speed, safety or cost benchmark has been published here. The project’s own test counts validate its stated release checks; they do not measure performance against ChatGPT, Gemini, Copilot, Claude or Letta. Local operation also brings responsibility for hardware, updates and backups.
