Recall
An operator-grade local memory and control runtime built around daemon authority, receipts, evidence, and repair handoffs.
Overview
Recall is the flagship runtime program: a local, evidence-first memory/control system whose differentiator is that it can testify. The goal is not just chat or retrieval. The goal is a system that can explain what happened, which evidence was used, what changed, which tool path executed, what failed, what was verified, and what a repair agent should do next.
Current Shape
The active workspace is Rust-first and split into dedicated components:
- recall-session: session state, provider config, agent policy, scheduler, governance, memory policy, and tests.
- recall-daemon: daemon runtime and IPC authority surface.
- recall-app: desktop/UI surface.
- recall-cli: command-line client.
- recall-ingest: document, chat, and file ingestion.
- recall-embedder: embedding and indexing support.
- recall-web: web/RSS/search support when explicitly enabled.
- recall-contracts and recall-ipc: typed contracts and client/daemon boundaries.
The authority model is daemon-owned writes. The app and CLI should act as clients, not peer runtimes.
Architecture Focus
Recall sits on the same evidence-runtime line as the Rust Libraries work:
- episode-first identity
- execution context as an artifact
- bitemporal truth
- query receipts
- repair records
- reference interpreters
- explicit plane ownership
- Codex packet generation for source-grounded repair passes
The library stack - semantic-memory, knowledge-runtime, llm-tool-runtime, stack-ids, and verification crates - is the substrate Recall should consume rather than reimplement app-locally.
Current Hardening Work
Recent audit material focuses on preventing scheduler, plan, and UI-storm classes of bugs before they become user-visible incidents. The main defect families are fail-open durability, whole-state persistence on hot paths, event-driven UI refetch amplification, and incomplete idempotency for child work units.
Next Gates
- Enforce daemon as the only write-capable authority.
- Generate runtime truth from backend state, not model text.
- Add machine-readable doctor reports.
- Gate web/current-info answers on actual tool availability.
- Emit receipts for every material operation.
- Produce Codex repair packets with source basis, issue matrix, doctor report, receipts, and verification plan.
Have questions about Recall?
Try asking the AI assistant! Here are some ideas:
Related Projects
Gloss
A local-first desktop knowledge application for grounded chat over personal documents, with notebook-local storage and source-linked answers.
VisionForge
A Tauri 2 app that connects Ollama and ComfyUI through a local prompt-engineering, queue, gallery, and generation-receipt workflow.
ClaimLedger
A claim and evidence compiler for Gloss Claim Hygiene, producing source-spanned, reviewable, deterministic bundles and testimony objects.