{"id":43952791,"url":"https://github.com/bug-ops/zeph","last_synced_at":"2026-04-07T04:01:23.345Z","repository":{"id":336739259,"uuid":"1150904280","full_name":"bug-ops/zeph","owner":"bug-ops","description":"Rust AI agent where every context token earns its place. Self-learning skills, temporal graph memory, cascade quality   routing, OWASP AI security. Hybrid inference: Ollama · Claude · Gemini · OpenAI · GGUF. MCP + ACP. One binary.","archived":false,"fork":false,"pushed_at":"2026-04-06T18:23:30.000Z","size":20619,"stargazers_count":19,"open_issues_count":21,"forks_count":3,"subscribers_count":2,"default_branch":"main","last_synced_at":"2026-04-06T18:26:04.296Z","etag":null,"topics":["ai-agent","claude","cli","gemini","graph-memory","huggingface","llm","local-llm","mcp","multi-model","ollama","openai","qdrant","rag","rust","self-learning","telegram-bot","tool-use","tree-sitter","tui"],"latest_commit_sha":null,"homepage":"https://bug-ops.github.io/zeph/","language":"Rust","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/bug-ops.png","metadata":{"files":{"readme":"README.md","changelog":"CHANGELOG.md","contributing":"CONTRIBUTING.md","funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":"SECURITY.md","support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":".github/CLA.md"}},"created_at":"2026-02-05T20:35:40.000Z","updated_at":"2026-04-06T18:19:21.000Z","dependencies_parsed_at":"2026-02-18T02:11:12.403Z","dependency_job_id":"7ef51906-842c-4128-adeb-35f96d45cba2","html_url":"https://github.com/bug-ops/zeph","commit_stats":null,"previous_names":["bug-ops/zeph"],"tags_count":58,"template":false,"template_full_name":null,"purl":"pkg:github/bug-ops/zeph","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bug-ops%2Fzeph","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bug-ops%2Fzeph/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bug-ops%2Fzeph/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bug-ops%2Fzeph/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/bug-ops","download_url":"https://codeload.github.com/bug-ops/zeph/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bug-ops%2Fzeph/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":31499193,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-04-07T03:10:19.677Z","status":"ssl_error","status_checked_at":"2026-04-07T03:10:13.982Z","response_time":105,"last_error":"SSL_connect returned=1 errno=0 peeraddr=140.82.121.5:443 state=error: unexpected eof while reading","robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":false,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["ai-agent","claude","cli","gemini","graph-memory","huggingface","llm","local-llm","mcp","multi-model","ollama","openai","qdrant","rag","rust","self-learning","telegram-bot","tool-use","tree-sitter","tui"],"created_at":"2026-02-07T04:03:56.211Z","updated_at":"2026-04-07T04:01:23.334Z","avatar_url":"https://github.com/bug-ops.png","language":"Rust","readme":"\u003cdiv align=\"center\"\u003e\n  \u003cimg src=\"book/src/assets/zeph_v8_github.png\" alt=\"Zeph\" width=\"800\"\u003e\n\n  **A Rust AI agent that learns from every session and remembers the reasoning behind every decision.**\n\n  [![Crates.io](https://img.shields.io/crates/v/zeph)](https://crates.io/crates/zeph)\n  [![docs](https://img.shields.io/badge/docs-book-blue)](https://bug-ops.github.io/zeph/)\n  [![CI](https://img.shields.io/github/actions/workflow/status/bug-ops/zeph/ci.yml?branch=main\u0026label=CI)](https://github.com/bug-ops/zeph/actions)\n  [![Tests](https://img.shields.io/badge/tests-7720-brightgreen)](https://github.com/bug-ops/zeph/actions)\n  [![codecov](https://codecov.io/gh/bug-ops/zeph/graph/badge.svg?token=S5O0GR9U6G)](https://codecov.io/gh/bug-ops/zeph)\n  [![Crates](https://img.shields.io/badge/crates-21-orange)](https://github.com/bug-ops/zeph/tree/main/crates)\n  [![MSRV](https://img.shields.io/badge/MSRV-1.88-blue)](https://www.rust-lang.org)\n  [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)\n\u003c/div\u003e\n\n---\n\n## Why Zeph\n\n- **Gets smarter with every task** — Wilson score Bayesian confidence tracks which skills actually work in practice. Underperforming skills lose priority; successful ones surface first. When repeated failures cluster around a skill, the agent generates an improved version autonomously — no configuration required.\n- **Remembers why, not just what** — Five-typed graph memory (Causal, Temporal, Semantic, CoOccurrence, Hierarchical) with SYNAPSE spreading activation. \"Why did we choose PostgreSQL?\" traverses causal edges through the decision graph — not keyword search through chat logs.\n- **Preserves working memory mid-task** — HiAgent subgoal-aware compaction identifies the current task goal before evicting context. Unlike FIFO trimming, information relevant to the active subgoal is never dropped mid-execution.\n- **Routes smarter, spends less** — declare providers once, route by complexity tier (Simple/Medium/Complex/Expert). Thompson Sampling and LinUCB bandit learn which provider wins per query type. Plan template caching reuses successful DAG plans to cut repeated-task cost.\n- **Security-first architecture** — age-encrypted vault for all secrets, 17-pattern MCP injection detection, OAP tool authorization, per-session tool quota, SSRF guards, and exfiltration detection — built into the core, not bolted on.\n\n\n---\n\n## Quick Start\n\n```bash\ncargo install zeph\nzeph init          # interactive setup wizard — picks provider, model, features\nzeph               # start the agent\n```\n\n**Tip:** Pre-built binaries and Docker images are on the [releases page](https://github.com/bug-ops/zeph/releases). `curl -fsSL https://github.com/bug-ops/zeph/releases/latest/download/install.sh | sh` also works.\n\n**Tip:** Copy-paste configs for all common setups — local, cloud, hybrid, coding assistant, Telegram bot — are in the **[Configuration Recipes](https://bug-ops.github.io/zeph/guides/config-recipes.html)** guide.\n\n---\n\n## Feature Highlights\n\n- [x] **[Self-learning skills](https://bug-ops.github.io/zeph/advanced/self-learning.html)** — Agent-as-a-Judge feedback detection (fast regex path + rate-limited LLM path), Wilson score Bayesian ranking promotes skills that actually work, autonomous skill evolution triggered by clustered failures, RL-based SleepGate admission control prevents noise from polluting long-term memory\n- [x] **[Graph memory with SYNAPSE](https://bug-ops.github.io/zeph/concepts/graph-memory.html)** — five typed edge categories (Causal, Temporal, Semantic, CoOccurrence, Hierarchical) via MAGMA; spreading activation retrieval with hop-by-hop decay and lateral inhibition surfaces multi-hop connections; community detection clusters entities by topic; BFS recall injected alongside vector results each turn\n- [x] **[Skills-first architecture](https://bug-ops.github.io/zeph/concepts/skills.html)** — YAML+Markdown skill files, hot-reload on edit, BM25+cosine hybrid retrieval with RRF fusion, Bayesian re-ranking\n- [x] **[Context engineering](https://bug-ops.github.io/zeph/advanced/context.html)** — three-tier compaction pipeline, HiAgent subgoal-aware eviction, failure-driven compression guidelines (ACON, ICLR 2026), Memex tool-output archival\n- [x] **[Semantic memory](https://bug-ops.github.io/zeph/concepts/memory.html)** — SQLite or PostgreSQL + Qdrant, MMR re-ranking, temporal decay, semantic response cache\n- [x] **[Multi-model orchestration](https://bug-ops.github.io/zeph/advanced/complexity-triage.html)** — complexity triage routing (Simple/Medium/Complex/Expert), Thompson Sampling, cascade cost tiers, PILOT LinUCB bandit\n- [x] **[Hybrid inference](https://bug-ops.github.io/zeph/concepts/providers.html)** — Ollama, Claude, OpenAI, Gemini, any OpenAI-compatible API, or fully local via Candle (GGUF)\n- [x] **[Task orchestration](https://bug-ops.github.io/zeph/concepts/task-orchestration.html)** — DAG-based task graphs with LLM goal decomposition, parallel execution, plan template caching\n- [x] **[MCP client](https://bug-ops.github.io/zeph/guides/mcp.html)** — full tool exposure, OAuth 2.1 + PKCE for remote servers, 17-pattern injection detection, per-session tool quota, OAP authorization\n- [x] **[Security sandbox](https://bug-ops.github.io/zeph/reference/security.html)** — age-encrypted vault, shell sandbox, file read sandbox, SSRF protection, PII filter, exfiltration guards\n- [x] **[ACP server](https://bug-ops.github.io/zeph/advanced/acp.html)** — stdio, HTTP+SSE, WebSocket transports for IDE integration (Zed, VS Code, Helix)\n- [x] **[A2A protocol](https://bug-ops.github.io/zeph/advanced/a2a.html)** — agent-to-agent delegation over JSON-RPC 2.0 with IBCT capability tokens\n- [x] **[Sub-agents](https://bug-ops.github.io/zeph/advanced/sub-agents.html)** — isolated agents with scoped tools, zero-trust secret delegation, persistent transcripts\n- [x] **[TUI dashboard](https://bug-ops.github.io/zeph/advanced/tui.html)** — ratatui-based with real-time metrics, security panel, plan view, command palette\n- [x] **[Multi-channel I/O](https://bug-ops.github.io/zeph/advanced/channels.html)** — CLI, Telegram, TUI, Discord, Slack — all with streaming, voice, and vision input\n- [x] **[LSP integration](https://bug-ops.github.io/zeph/guides/lsp.html)** — compiler-level code intelligence via rust-analyzer, pyright, gopls and others: type info, diagnostics, call hierarchy, safe rename, references — injected automatically into context after file writes and reads\n- [x] **[Code indexing](https://bug-ops.github.io/zeph/advanced/code-indexing.html)** — tree-sitter AST-based indexing (Rust, Python, JS, TS, Go), semantic search, repo map generation\n- [x] **[Document RAG](https://bug-ops.github.io/zeph/advanced/document-loaders.html)** — ingest `.txt`, `.md`, `.pdf` into Qdrant with automatic retrieval per turn\n- [x] **[Self-experimentation](https://bug-ops.github.io/zeph/concepts/experiments.html)** — autonomous LLM config tuning via grid sweep, random sampling, neighborhood search\n- [x] **[Config migration](https://bug-ops.github.io/zeph/guides/migrate-config.html)** — `zeph migrate-config --diff` previews and applies config upgrades after version bumps\n- [x] **Single binary** -- ~15 MB, no runtime dependencies, ~50 ms startup, ~20 MB idle memory\n\n---\n\n## Architecture\n\n```text\nzeph (binary)\n |\n +-- zeph-core            agent loop, context builder, metrics, channel trait\n |    |\n |    +-- zeph-config     TOML config, env overrides, migration, init wizard\n |    +-- zeph-db         SQLite/PostgreSQL pool, migrations, store trait\n |    +-- zeph-vault      age-encrypted secret storage, vault resolution\n |    +-- zeph-common     shared types, error utilities, tracing helpers\n |    +-- zeph-sanitizer  content sanitization, injection detection, PII filter\n |    |\n |    +-- zeph-llm        LlmProvider trait, Ollama/Claude/OpenAI/Gemini/Candle backends\n |    +-- zeph-skills     SKILL.md parser, registry, embedding matcher, self-learning\n |    +-- zeph-memory     semantic memory orchestrator, graph memory, SYNAPSE\n |    +-- zeph-tools      ToolExecutor trait, shell/web/file/composite executors, audit\n |    +-- zeph-mcp        MCP client, multi-server lifecycle, tool registry\n |    +-- zeph-orchestration  DAG task graphs, planner, scheduler, aggregator\n |    +-- zeph-subagent   sub-agent spawner, transcript persistence, lifecycle hooks\n |    +-- zeph-index      AST code indexing, semantic retrieval, repo map\n |\n +-- zeph-channels        CLI, Telegram, Discord, Slack adapters\n +-- zeph-tui             ratatui TUI dashboard (feature-gated)\n +-- zeph-acp             ACP server: stdio/HTTP+SSE/WebSocket (feature-gated)\n +-- zeph-a2a             A2A client + server, agent discovery (feature-gated)\n +-- zeph-gateway         HTTP webhook gateway with bearer auth (feature-gated)\n +-- zeph-scheduler       cron-based periodic tasks (feature-gated)\n +-- zeph-experiments     autonomous LLM config experimentation engine\n```\n\nOptional features are grouped into use-case bundles: `desktop` (TUI), `ide` (ACP), `server` (gateway + A2A + otel), `chat` (Discord + Slack), `ml` (Candle + PDF). Use `--features full` for everything except hardware-specific GPU flags. See [Feature Flags](https://bug-ops.github.io/zeph/reference/feature-flags.html).\n\n---\n\n```text\n┌─ Skills (3/12) ────────────────────┐┌─ MCP Tools ─────────────────────────┐\n│  web-search  [████████░░] 82% (117)││  - filesystem/read_file             │\n│  git-commit  [███████░░░] 73%  (42)││  - filesystem/write_file            │\n│  code-review [████░░░░░░] 41%   (8)││  - github/create_pr                 │\n└────────────────────────────────────┘└─────────────────────────────────────┘\n```\n\n\u003cdiv align=\"center\"\u003e\n  \u003cimg src=\"book/src/assets/zeph.gif\" alt=\"Zeph TUI Dashboard\" width=\"800\"\u003e\n\u003c/div\u003e\n\n## Documentation\n\nFull documentation — installation, configuration, guides, and architecture reference — at **[bug-ops.github.io/zeph](https://bug-ops.github.io/zeph/)**.\n\nZeph's design draws from a broad range of published research: parallel tool execution ([LLMCompiler, ICML 2024](https://arxiv.org/abs/2312.04511)), failure-driven context compression ([ACON, ICLR 2026](https://arxiv.org/abs/2510.00615)), temporal knowledge graphs ([Zep/Graphiti, 2025](https://arxiv.org/abs/2501.13956)), agentic memory linking ([A-MEM, NeurIPS 2025](https://arxiv.org/abs/2502.12110)), observation masking and schema-based summarization ([Manus, 2025](https://rlancemartin.github.io/2025/10/15/manus/)), and more. The full list of papers, blog posts, and specifications that shaped Zeph is at **[References \u0026 Inspirations](https://bug-ops.github.io/zeph/references.html)**.\n\n## Contributing\n\nSee [CONTRIBUTING.md](CONTRIBUTING.md). Found a vulnerability? Use [GitHub Security Advisories](https://github.com/bug-ops/zeph/security/advisories/new).\n\n## License\n\n[MIT](LICENSE)\n","funding_links":[],"categories":[],"sub_categories":[],"project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fbug-ops%2Fzeph","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fbug-ops%2Fzeph","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fbug-ops%2Fzeph/lists"}