Kiro-Ception adds persistent cross-machine memory to Kiro agent
Farley Farley built Kiro-Ception, an open-source Kiro Power that indexes all CLI and IDE conversation history locally and gives the Kiro agent automatic memory recall across sessions, projects, and multiple machines.
Score breakdown
Kiro-Ception fills the gap left by Kiro's lack of native persistent memory, giving the agent automatic recall of past conversations across all projects, sessions, and machines without any data leaving the user's machine by default.
- 01Built by Farley Farley after 7 months of daily Kiro use across 100+ projects and multiple computers
- 02Installed as a Kiro Power with keyword triggers that auto-fire on natural language cues like 'as we discussed' or 'remind me'
- 03Hybrid search combines semantic embeddings (meaning) and FTS5 keyword matching (exact identifiers)
Farley Farley built Kiro-Ception after seven months of daily Kiro use across more than 100 projects and multiple computers, frustrated by the agent's inability to recall past work — especially across machines. The tool is installed as a Kiro Power and indexes all Kiro CLI and IDE conversation history in the background, starting with the newest conversations first. Its hybrid search engine combines semantic embeddings (for meaning-based retrieval) with FTS5 keyword search (for exact function names and identifiers), delivering search results in under 10ms with cold-starts under 2 seconds from cache.
The default embedding model is approximately 80MB and downloads once; GPU-based higher-end embedding models are also supported.
By default, Kiro-Ception runs entirely locally using CPU-based embeddings, SQLite, and numpy. The default embedding model is approximately 80MB and downloads once; GPU-based higher-end embedding models are also supported. Users can optionally plug in Ollama or OpenAI for more powerful models. An optional multi-node search feature allows sessions to be found across multiple computers using encrypted transport. The project is fully open-source under the MIT license, with zero telemetry and zero external calls. Setup involves cloning the repo, running `uv sync`, and adding the Power via its local path in Kiro.
Key facts
- 01Built by Farley Farley after 7 months of daily Kiro use across 100+ projects and multiple computers
- 02Installed as a Kiro Power with keyword triggers that auto-fire on natural language cues like 'as we discussed' or 'remind me'
- 03Hybrid search combines semantic embeddings (meaning) and FTS5 keyword matching (exact identifiers)
- 04Search results return in under 10ms; cold-starts are under 2 seconds from cache
- 05Runs 100% locally by default using CPU-based embeddings, SQLite, and numpy — zero telemetry, zero external calls
- 06Default embedding model is ~80MB and downloads once; Ollama or OpenAI can be substituted for higher-end models
- 07Optional multi-node mode enables encrypted cross-machine session search; fully open-source under MIT license
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 18, 2026 · 10:40 UTC. How this works →