Sallah Kokaina

OSS radar

Open-source projects that caught my curiosity. The ones I've tried get a rating and a verdict, with my notes and any article or project that came out of it. The rest are on my radar, yet to experiment.

jump to:tried & ratedagent memory & retrievalknowledge graphs & ontologiescoding agents: orchestration & toolingagent platforms & harnessesagent security & sandboxingdecision models & evalsdocument parsing & extractionlocal models & self-hostingenterprise infrastructure for agentsenterprise harness

Tried & rated16 projects I've evaluated, most starred first.

01Tried & rated

16

On my radar123 projects that caught my curiosity, trending or on my list to try. Most starred first.

02Agent memory & retrieval

19

Long-term memory, LLM wikis and retrieval layers for agents.

  • OpenVikingyet to experimentvolcengine/OpenViking★ 41.6k

    Context database exposing memory, resources and skills as a viking:// filesystem.

  • TencentDB Agent Memoryyet to experimentTencentCloud/TencentDB-Agent-Memory★ 30.5k

    Tencent's layered, fully local long-term agent memory on SQLite.

  • EverOSyet to experimentEverMind-AI/EverOS★ 13.8k

    Self-hosted memory tier with plain Markdown on disk as the source of truth.

  • MemOSyet to experimentMemTensor/MemOS★ 12.3k

    Memory operating system for LLMs and agents, including learned behaviour.

  • Wigoloyet to experimentKnockOutEZ/wigolo★ 5.5k

    Local web search, fetch and crawl for agents across 18 engines, no API keys.

  • OpenKByet to experimentVectifyAI/OpenKB★ 4.8k

    Compiles documents into an LLM-maintained wiki, with vectorless PageIndex retrieval.

  • zvec-grepyet to experimentzvec-ai/zvec-grep★ 4.2k

    Semantic code search CLI fusing ripgrep, BM25 and vectors.

  • EdgeQuakeyet to experimentraphaelmansuy/edgequake★ 2.3k

    Graph-RAG in Rust on a single Postgres, cloud or local LLMs.

  • OmniGraphyet to experimentModernRelay/omnigraph★ 1.9k

    Rust graph database with Git-style branching and merging for shared agent memory.

  • Leviathanyet to experimentelstongun/leviathan★ 879

    Keyword search engine that lets agents query million-row datasets cheaply.

  • OKF Agent Memoryyet to experimentokf-memory/okf-agent-memory★ 852

    Git-native agent memory on Google's Open Knowledge Format, served over MCP.

  • LatticeDByet to experimentjeffhajewski/latticedb★ 746

    Embedded single-file database for graph, vector and full-text search.

  • CortexDByet to experimentliliang-cn/cortexdb★ 475

    Agent memory and knowledge graph in one SQLite file, pure Go, with MCP.

  • exxpertsyet to experimentEXXETA/exxperts★ 444

    Local-first AI colleague per project with approval-gated persistent memory.

  • llm-wiki-v2yet to experimentcrow-hq/llm-wiki-v2★ 255

    Scalable LLM wiki engine using the CROW architecture (AGPL-3.0).

  • Wifeyet to experimentma-nucho-pro/wife★ 69

    Local, auditable memory layer for Claude Code, Codex and other agents.

  • Crow Desktopyet to experimentcrow-hq/crow-desktop★ 18

    Desktop app that turns your local documents into a chat-able LLM wiki.

  • dejayet to experimentCynepMyx/deja★ 1

    Indexes past coding-agent session logs and serves search over MCP.

  • Oracle AI Agent Memoryyet to experimentoracleagentmemory

    Graph-aware, database-backed memory framework for agents (PyPI package).

03Knowledge graphs & ontologies

13

Semantic layers and ontologies that ground agents in business data.

  • Semanticayet to experimentsemantica-agi/semantica★ 14.4k

    Records agent decisions as graph nodes with rationale and provenance.

  • Utopiayet to experimentdeeplethe/utopia★ 8.7k

    Self-hosted enterprise world model: a dated knowledge graph built from documents.

  • knowledge_graphyet to experimentrahulnyk/knowledge_graph★ 4.2k

    Concept graph from text with a local 7B model, visualised with Pyvis.

  • open-ontologiesyet to experimentfabio-rovai/open-ontologies★ 2.8k

    Terraform-style plan and apply for ontology changes, over MCP.

  • Helix Foundryyet to experimentHelixDB/helix-foundry★ 2.1k

    Local-first ontology builder across your databases and SaaS tools, on HelixDB.

  • GraphGenyet to experimentInternScience/GraphGen★ 1.2k

    Builds knowledge graphs from documents to generate verifiable training data.

  • EvoOntologyyet to experimentruc-datalab/EvoOntology★ 1.2k

    Self-evolving ontology for data agents, packaged as an MCP server.

  • context-ontology-acceleratoryet to experimentaws/context-ontology-accelerator★ 1.1k

    AWS semantic layer that induces an ontology and serves it over MCP or SPARQL.

  • Docling Graphyet to experimentdocling-project/docling-graph★ 937

    Builds typed knowledge graphs from PDFs and Office files via Pydantic templates.

  • SLayeryet to experimentMotleyAI/slayer★ 334

    Semantic layer for agents so metrics are defined once, not re-guessed in SQL.

  • OntoCastyet to experimentgrowgraph/ontocast★ 253

    Extracts RDF triples from documents and evolves the ontology in the same run.

  • OrionBeltyet to experimentralforion/orionbelt-ontology-builder★ 221

    Browser workbench for editing OWL ontologies and SKOS vocabularies.

  • Ontology Atlasyet to experimentwlsdks/ontology-atlas★ 149

    Codebase ontology as Markdown in a folder, readable and writable by agents.

04Coding agents: orchestration & tooling

27

Running many coding agents in parallel, keeping their context lean, and checking their work.

  • REAyet to experimentmorluto/rea★ 434k

    Gives coding agents Ghidra-backed reverse engineering of binaries over MCP.

  • Diagram Designyet to experimentcathrynlavery/diagram-design★ 65.3k

    Agent skill that answers with diagrams across 28 diagram types.

  • Open Code Reviewyet to experimentalibaba/open-code-review★ 49.5k

    Alibaba's code reviewer combining deterministic rules with an LLM agent.

  • Context Modeyet to experimentmksglu/context-mode★ 33.6k

    Sandboxes large tool output so only the answer enters the agent's context.

  • Bendyet to experimentbendlang/bend★ 24.8k

    Language where agents must prove code satisfies declared laws before merging.

  • Paseoyet to experimentgetpaseo/paseo★ 23.2k

    Local daemon to drive coding-agent sessions from desktop, phone, web or CLI.

  • Plandexyet to experimentplandex-ai/plandex★ 15.9k

    Terminal coding agent built for large projects and multi-file tasks.

  • e2eyet to experimenttester-army/e2e★ 13.3k

    Agent-recorded end-to-end tests for web and mobile, replayed without model calls.

  • Worktrunkyet to experimentmax-sixty/worktrunk★ 11.4k

    A git worktree per parallel agent, with a shared build cache.

  • Autoharnessyet to experimenttigerless-labs/autoharness★ 10.7k

    Distils and maintains Claude Code skills from your real sessions.

  • OpenSpaceyet to experimentHKUDS/OpenSpace★ 7.8k

    Scores and evolves agent skills based on real task outcomes.

  • TeamAI-CLIyet to experimentTencent/teamai-cli★ 5.8k

    Shares a team's skills, rules and hooks across agents from one git repo.

  • Weave Routeryet to experimentweave-os/router★ 5.6k

    Routes each coding-agent request to the best available model.

  • forkdyet to experimentdeeplethe/forkd★ 4.8k

    Forks live Firecracker microVMs so an agent can branch its plan mid-run.

  • Zeronyet to experimentzeronsh/zeron★ 3.5k

    Native control plane for Claude Code, Codex, Cursor, Devin and other coding agents.

  • Zeroshotyet to experimentthe-open-engine/zeroshot★ 2.1k

    Pairs every agent change with an independent verifier that never saw how it was written.

  • Codyncyet to experimentleepokai/Codync★ 1.9k

    Turns local coding agents into persistent bots you can message from anywhere.

  • Superlogyet to experimentsuperloglabs/superlog★ 1.5k

    OpenTelemetry incidents exposed to coding agents through MCP tools.

  • Pullfrogyet to experimentpullfrog/pullfrog★ 1.4k

    Bring-your-own-key GitHub bot running Claude Code, Codex or OpenCode in Actions.

  • gitaiyet to experimentfeder-cr/git-ai★ 1k

    Git server for agents based on Anthropic's Scion design: merge first, test second.

  • old-coderyet to experimentAmazingAng/old-coder★ 757

    Skill that makes agents pass a test gauntlet instead of you reading the diff.

  • agentacctyet to experimentmikehasa/agentacct★ 755

    Turns coding-agent logs into work receipts to check what was really done.

  • Speculayet to experimentspecula-org/Specula★ 629

    Coding agents write TLA+ specs and model-check code for concurrency bugs.

  • Boostyet to experimentjfrog/boost★ 545

    Compresses noisy command output into structured context for agents.

  • Vicoayet to experimentvicoa-ai/vicoa★ 531

    Self-hostable orchestrator for teams of coding agents across 40+ tools.

  • OrcaReplayyet to experimentContinuum-AI-Corp/OrcaReplay★ 345

    Records agent-model traffic so runs can be replayed or forked offline.

  • Kanderyet to experimentdualface/kander★ 172

    Terminal kanban that schedules parallel coding agents with review gates.

05Agent platforms & harnesses

18

Harnesses, runtimes and self-hosted platforms for running agents.

  • DeepSeek Harnessyet to experimentdeepseek-ai/deepseek-harness★ 274k

    Agent harness where the model adapter, tools and loop are all swappable plugins.

  • Lightpandayet to experimentlightpanda-io/browser★ 41.4k

    Lightweight headless browser for AI; agent sessions export as replayable scripts.

  • Obscurayet to experimenth4ckf0r0day/obscura★ 30.4k

    Fast Rust headless browser engine for scraping and agents.

  • Haystackyet to experimentdeepset-ai/haystack★ 26.9k

    Explicit, controllable pipelines for RAG and agent workflows.

  • Jev Ultrafastyet to experimentbrowser-use/jev-ultrafast★ 24.2k

    Browser agent that reads the element table instead of screenshots.

  • OpenWorkeryet to experimentandrewyng/openworker★ 18.9k

    Andrew Ng's local-first desktop AI coworker with approval gates.

  • ARTEMISyet to experimentgoogle/artemis★ 16.7k

    Google agent that operates real Android phones from natural language.

  • OpenSREyet to experimentTracer-Cloud/opensre★ 12.7k

    Framework for AI SRE agents that investigate incidents with evidence.

  • Octopyet to experimentTencentCloud/Octop★ 11.7k

    Tencent's self-hosted, multi-user assistant with a workspace per agent.

  • OpenBotyet to experimentCopilotKit/OpenBot★ 10.2k

    Self-hosted AI coworkers with real tool and browser access.

  • TrueForgeyet to experimenttruefoundry/trueforge★ 6.4k

    Agent harness giving teams a chat UI and apps an API, with sandboxed code.

  • Apache Makayet to experimentapache/maka★ 5.9k

    Local-first agent workspace with an append-only event log (Apache incubator).

  • Awesome Harness Engineeringyet to experimentai-boost/awesome-harness-engineering★ 5.1k

    Curated resources and patterns for building reliable agent harnesses.

  • Switchyet to experimentsandbox-quantum/switch★ 3.9k

    Puts agents and people in the same chat channel, no pre-wiring.

  • NOOAyet to experimentNVIDIA-NeMo/labs-OO-Agents★ 2.3k

    NVIDIA's framework that defines agents as plain Python objects.

  • Agent Mesh (SAM)yet to experimentgoogle/agentmesh★ 1.2k

    Google's peer-to-peer mesh that lets agents discover and share tools.

  • HugAgentOSyet to experimentZJU-REAL/HugAgentOS★ 628

    Checks every agent plan against rules, pausing risky steps for approval.

  • Platypusyet to experimentwilldady/platypus★ 130

    Self-hosted full-stack app for teams building agents on their own infra.

06Agent security & sandboxing

9

Guardrails, firewalls and isolated runtimes for agents that run real commands.

  • CubeSandboxyet to experimentTencentCloud/CubeSandbox★ 13.2k

    Hardware-isolated, E2B-compatible agent sandboxes with sub-60ms start.

  • FailproofAIyet to experimentFailproofAI/failproofai★ 8.1k

    Policy hooks that block risky agent tool calls before they run.

  • AI-Infra-Guardyet to experimentTencent/AI-Infra-Guard★ 7.4k

    Tencent's scanner for agent workflows, MCP servers, skills and model jailbreaks.

  • Agent Sandboxyet to experimentkubernetes-sigs/agent-sandbox★ 4.5k

    Kubernetes CRD for stateful agent pods with persistent storage and identity.

  • ArcBoxyet to experimentarcboxlabs/arcbox★ 4.3k

    Runs Claude Code in its own Firecracker microVM instead of approval clicks.

  • Agent Beaconyet to experimentAsymptote-Labs/agent-beacon★ 3.3k

    Records agent actions across 23+ harnesses and turns good runs into shared skills.

  • ADRyet to experimentuber/ADR★ 2.1k

    Uber's agent detection and response, built from Claude Code and Codex logs.

  • HOL Guardyet to experimenthashgraph-online/hol-guard★ 1.1k

    Local-first firewall for agent tools, MCP servers, skills and packages.

  • Pipelockyet to experimentpipelab-org/pipelock★ 984

    Inspects HTTP, MCP and A2A traffic for exfiltration and injection.

07Decision models & evals

12

System-1 decision models, and benchmarks and evals for agents.

  • DeepEvalyet to experimentconfident-ai/deepeval★ 19.3k

    LLM eval framework, now with a Jev-as-judge metric (JevEval).

  • OpenEnvyet to experimenthuggingface/OpenEnv★ 2.8k

    Isolated, Gymnasium-style environments for agentic RL.

  • Clefyet to experimentCloudflare/clef♥ 2.3k

    Cloudflare's open-weight decision models (Clef and Clef-flash), Apache-2.0.

  • Wazayet to experimentmicrosoft/waza★ 1.5k

    Scaffolds and runs eval suites for SKILL.md agent skills.

  • pg-jevyet to experimentrealZachi/pg-jev★ 1.3k

    Postgres extension for semantic filtering and ranking in SQL via Jev.

  • CLMyet to experimentContrastive-LM/CLM-v0.1-8B♥ 790

    Scores candidate actions in embedding space for routing and verification.

  • Strands Decideryet to experimentstrands-labs/strands-decider★ 764

    Open decision model that scores options in one forward pass.

  • Jevboxyet to experimentextend-hq/jevbox★ 756

    Self-organising document drive that files, answers and cites, powered by Jev.

  • Abideyet to experimentcoldteadotai/abide★ 623

    Enforces AGENTS.md rules on every edit with fast Jev checks.

  • GLiNER2.5-Decideyet to experimentfastino/GLiNER2.5-Decide♥ 451

    340M encoder for decisions, spans and relations that runs on CPU.

  • SWE-sweepyet to experimentfacebookresearch/swe-sweep★ 222

    Meta benchmark: agents must find and fix bugs with no bug report.

  • Tev1-4B-experimentalyet to experimenttogethercomputer/Tev1-4B-experimental♥ 38

    Together AI fine-tune that picks among 2 to 24 options.

08Document parsing & extraction

7

OCR, parsing and structured extraction from PDFs, scans and web pages.

  • Markeryet to experimentdatalab-to/marker★ 40.8k

    Converts PDFs, images, docs and slides into clean Markdown.

  • ScrapeGraphAIyet to experimentScrapeGraphAI/Scrapegraph-ai★ 32.5k

    LLM-powered scraping of web pages into structured JSON.

  • docTRyet to experimentmindee/doctr★ 6.4k

    OCR on your own hardware, with every word's box and confidence.

  • TurboOCRyet to experimentaiptimizer/TurboOCR★ 1.2k

    GPU document parser returning text, layout, HTML tables and LaTeX.

  • receipt-ocryet to experimentbhimrazy/receipt-ocr★ 753

    Receipt extraction to your own JSON schema with any model provider.

  • fenicyet to experimenttypedef-ai/fenic★ 679

    DataFrame engine with semantic extract into typed Pydantic fields.

  • WeMM-Embeddingyet to experimenttencent/WeMM-Embedding-2B♥ 125

    Tencent's 2B embedding model for text, images and video in one space.

09Local models & self-hosting

4

Running models and apps on your own hardware.

  • Unslothyet to experimentunslothai/unsloth★ 79.1k

    Fast, low-memory fine-tuning and quantisation of open models.

  • LocalAIyet to experimentmudler/LocalAI★ 50.2k

    OpenAI-compatible local API for text, vision, voice and image models.

  • Openshipyet to experimentoblien/openship★ 15.2k

    Self-hosted deployment platform: repo in, app with domain and TLS out.

  • Magnitudeyet to experimentmagnitudedev/magnitude★ 7.7k

    Profiles your machine and ranks the models it can run.

10Enterprise infrastructure for agents

8

Streaming, durable execution, data catalogs, authorization, identity, networking and gateways for agents in production.

  • NetBirdyet to experimentnetbirdio/netbird★ 30.7k

    Zero-trust network over peer-to-peer WireGuard, connecting people, machines and agents.

  • NATSyet to experimentnats-io/nats-server★ 21k

    Cloud and edge messaging system for services and agents.

  • ZITADELyet to experimentzitadel/zitadel★ 15.5k

    Identity infrastructure: SSO, OIDC, SAML and multi-tenancy.

  • RisingWaveyet to experimentrisingwavelabs/risingwave★ 9.4k

    Event streaming platform that ingests, transforms and serves real-time data to agents.

  • Bifrostyet to experimentmaximhq/bifrost★ 9.4k

    LLM, MCP and agent gateway. The open-source tier includes virtual keys for access control, hierarchical budgets and per-consumer rate limits.

  • OpenFGAyet to experimentopenfga/openfga★ 6.1k

    Fine-grained authorization engine inspired by Google Zanzibar.

  • Restateyet to experimentrestatedev/restate★ 4.7k

    Durable execution for resilient workflows and agents that survive infrastructure failures (BSL 1.1 license).

  • Lakekeeperyet to experimentlakekeeper/lakekeeper★ 1.5k

    Apache Iceberg REST catalog in Rust, with access control, credential vending and audit for engines and agents.

11Enterprise harness

6

Platforms that run, govern and share agents, skills and knowledge across an organization.

  • OpenHumanyet to experimenttinyhumansai/openhuman★ 42.6k

    Lightweight open-source agent harness that runs hundreds of agents on a small VPS.

  • Onyxyet to experimentonyx-dot-app/onyx★ 33.2k

    Open-source AI platform: chat, search and agents over company knowledge, with any LLM.

  • Cogneeyet to experimenttopoteretes/cognee★ 33.1k

    Open-source memory platform giving agents persistent long-term memory.

  • SkillHubyet to experimentiflytek/skillhub★ 5.4k

    Self-hosted agent skill registry with versioning, RBAC and audit logs.

  • TrustGraphyet to experimenttrustgraph-ai/trustgraph★ 2.9k

    Semantic agent governance platform built on knowledge graphs.

  • Evotownyet to experimentEXboys/evotown★ 157

    Governance layer for enterprise agent runtimes: skills, knowledge, observability, review and private distribution.