continuum-recall 0.1.0

A memory system that prevents repeated failures by learning from past mistakes.

Status: v0.1.0 beta — the package is installable and tested, but the state schema may still change before v1.0.

Continuum Memory

Memory should prevent failure.

Continuum is a structured memory system for AI coding agents. It records project decisions, constraints, rejected paths, lessons, and tests so an agent can avoid repeating known mistakes. It is not a general-purpose knowledge base, a replacement for project documentation, or a raw chat-history store.

The core rule is: the model proposes, the system verifies. Reviewer models can suggest memory candidates, but durable state is written only after deterministic validators check structure, evidence, specificity, sensitivity, and contradictions.

Install

pip install continuum-recall

The PyPI distribution name is continuum-recall. The Python import name is continuum.

Five-Minute Quickstart

This flow uses the default rule-based reviewer and does not require Ollama.

continuum init
continuum add "Decision: use AGPL-3.0-or-later"
continuum review
continuum state
continuum packet --task "write the README license section"

The packet includes the license decision as task-specific context and writes a copy under .continuum/packets/.

What Continuum Stores

Continuum creates a .continuum/ directory in your workspace:

• events.jsonl: append-only source events.
• state.json: validated continuity items.
• ignored.jsonl: rejected candidates and validator reasons.
• diffs/: review diffs.
• packets/: generated context packets.
• ablation/: causal impact results.
• index/: search indexes.
• logs/reviewer.log: reviewer and validator logs.

All durable IDs and timestamps are minted by storage code, not by reviewer models.

CLI Surface

Common commands:

Command	Description
continuum init	Initialize a workspace.
continuum add "<text>"	Append a manual event and run review.
continuum review	Run the review cycle.
continuum state	Show state summary or JSON.
continuum packet --task "<task>"	Build a ranked context packet.
continuum search "<query>"	Search active items.
continuum test	Run deterministic memory tests.
continuum ablate <id> --task "<task>"	Measure single-item causal impact.

Optional Ollama Reviewer

The default reviewer is deterministic and model-free. An Ollama-backed reviewer committee is available by configuration, but Ollama is optional and not required for the getting-started flow.

Documentation

The documentation site covers architecture, context packets, memory tests, ablation, ContinuumBench, prior art, licensing, CLI, and SDK usage.

• Documentation: https://gutsy03.github.io/continuum/
• Repository: https://github.com/gutsy03/continuum

License

Continuum Memory is licensed under AGPL-3.0-or-later, with a commercial dual-license option for organizations that cannot use AGPL software in closed-source or hosted products.