# contemplative-agent

> A self-improving AI agent that learns from experience. Runs entirely on a local 9B model. Security by absence — dangerous capabilities were never built.

- **URL**: https://www.freshcrate.ai/projects/contemplative-agent
- **Author**: shimo4228
- **Category**: Security
- **Latest version**: `v2.5.0` (2026-06-05)
- **License**: MIT
- **Source**: https://github.com/shimo4228/contemplative-agent
- **Homepage**: https://doi.org/10.5281/zenodo.19212119
- **Language**: Python
- **GitHub**: 4 stars, 1 forks
- **Registry**: github
- **Tags**: `agent-framework`, `agent-simulation`, `ai-agent`, `ai-ethics`, `ai-safety`, `ai-security`, `autonomous-agent`, `contemplative-ai`, `python`

## Description

A self-improving AI agent that learns from experience. Runs entirely on a local 9B model. Security by absence — dangerous capabilities were never built.

## Recent releases

| Version | Date | Urgency | Changes |
| --- | --- | --- | --- |
| `v2.5.0` | 2026-06-05 | High | This release lands the 2026-06-04 full-stack agent-architecture audit (findings fixed across all four severity bands) and the three-ADR arc the audit triggered. **[ADR-0050](docs/adr/0050-epistemic-taxonomy-and-approval-lineage.md)** makes the self-conditioning loop *observable* — a two-valued {observed, generated} epistemic kind derived at read time from `provenance.source_type`, plus approval lineage (`source_ids` + `epistemic_counts`) plumbed through every promotion-producing command into `a |
| `v2.4.0` | 2026-05-25 | High | Three strands converge in this release. **(1) Self-post pipeline repair** (ADR-0039 / 0041 / 0043 / 0044): the self-post path had drifted into a silent echo chamber — a boolean Jaccard gate calibrated against one bad week had collapsed the post rate to ~1/day, an `extract_topics` summariser was collapsing peer voices into the agent's own canon, and a redundant `{topic_keywords}` injection plus a dead search rotation were leaking canon vocabulary to Moltbook's logs. The gate becomes a continuous |
| `v2.3.0` | 2026-05-05 | High | Cleanup-and-converge release after v2.2.x. Three sunset ADRs (0034 / 0035 / 0036) retire the experimental paths that v2.0.0 introduced (memory evolution, BM25 hybrid retrieval, skill-as-memory loop) and consolidate the surviving helpers into a single shape. ADR-0037 records — descriptively, not prescriptively — that the memory subsystem has converged on the Yogācāra eight-consciousness frame already named in ADR-0017.  Net diff vs. v2.2.1: 110 files changed, +2170 / -5772 (-3602 LOC), test file |
| `v2.2.1` | 2026-04-30 | High | Same-day correction of v2.2.0. Documentation-only; no code change.  Two factual errors in [ADR-0033](docs/adr/0033-aap-quadrant-lens-usage-note.md) Observations were corrected after re-reading `core/stocktake.py`, `adapters/dialogue/peer.py`, and `adapters/meditation/{pomdp,meditate}.py`.  ## Fixed  - **`skill-stocktake` and `dialogue` placement.** v2.2.0 described both as sitting at the "LLM Workflow ↔ Autonomous Agentic Loop boundary". On code re-read both have fixed control flow + bounded LLM |
| `v2.1.0` | 2026-04-21 | High | ## Highlights  - **Pluggable generation backend** via the new `LLMBackend` Protocol (`core/llm.py`). Sanitization, circuit breaker, and untrusted-content wrapping stay in core; the backend slot is what the [`contemplative-agent-cloud`](https://github.com/shimo4228/contemplative-agent-cloud) add-on fills to run the same CLI through Claude or GPT without a local Ollama. Default behavior unchanged — no backend = built-in Ollama HTTP. - **Local two-agent dialogue** — `contemplative-agent dialogue HO |
| `v2.0.0` | 2026-04-16 | High | # v2.0.0 — Yogācāra Memory Architecture  This release overhauls the Layer 2 knowledge store. The old discrete-category classification is retired; patterns are now stored as embedding coordinates, carry provenance / bitemporal validity / retrieval-aware forgetting, and co-evolve with their neighbors. The Yogācāra eight-consciousness model is adopted as the explicit architectural frame ([ADR-0017](../docs/adr/0017-yogacara-eight-consciousness-frame.md)): episode log ↔ sense-streams, knowledge ↔ āl |
| `v1.3.1` | 2026-04-07 | High | ## ADR-0015: One External Adapter Per Agent  Codifies the principle that a single agent process owns at most one adapter producing externally-observable side effects. Workflows spanning multiple external surfaces (e.g. posting *and* payment) must be decomposed into separate agent processes with separated authority, rather than bolted onto one.  This mirrors decades of organizational practice — segregation of duties, four-eyes approval, least privilege, Unix philosophy — which the broader LLM age |
| `v1.3.0` | 2026-04-07 | Medium | ## Highlights  ### Deterministic Dedup Gates Two-layer dedup against runaway self-similarity in agent output. The existing LLM-based gates (`check_topic_novelty`, `score_relevance`) are probabilistic and were observed to let through 40 near-identical self-posts in 7 days during the weekly analysis window of 2026-04-05. v1.3.0 adds deterministic gates that **complement** (not replace) the LLM gates:  - **Self-post pipeline**: prefix-5 stem + Jaccard similarity gate over   `(title ∪ topic_summary) |
| `v1.2.0` | 2026-03-31 | Medium | ## Highlights  ### Weekly Analysis Reports `contemplative-agent install-schedule --weekly-analysis` generates automated weekly behavioral analysis using `claude -p` via launchd. Reports track identity drift, knowledge growth, and constitutional evolution over 7-day windows.  ### Deduplication Scalability Knowledge store deduplication now uses importance-score decay filtering, preventing unbounded growth as episode logs accumulate. LLM parse robustness improved with code fence stripping and comma |
| `v1.1.0` | 2026-03-29 | Medium | ## What's New  ### Template Selection for Init `contemplative-agent init --template stoic` — choose from 10 ethical framework templates (Contemplative, Stoic, Utilitarian, Deontologist, Care Ethicist, Pragmatist, Existentialist, Contractarian, Cynic, Narrativist). Copies identity, constitution, skills, and rules from the selected template.  ### Importance Scoring Stability Fixed silent fallback to 0.5 when the 9B model returns malformed JSON. Now strips code fences, falls back to comma-separated |

## Citation

- HTML: https://www.freshcrate.ai/projects/contemplative-agent
- Markdown: https://www.freshcrate.ai/projects/contemplative-agent.md
- Dependencies JSON: https://www.freshcrate.ai/api/projects/contemplative-agent/deps

_Generated by freshcrate.ai. Indexes github releases for AI-agent ecosystem packages._