Open Source Agent Tools: November 12, 2026
Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled
1. Neo4j Releases Relate, an Apache 2.0 Embedded Graph Database in Rust Aimed at Agent Sandboxes and Agent Memory
Neo4j released Relate on November 11, an in-memory, in-process embedded graph database written in Rust and published as open source under the Apache 2.0 license. It ships with a built-in MCP server and a CLI, and exposes Bolt and HTTP Query API v2 endpoints so existing graph drivers and orchestration frameworks can connect without changes. Several behaviors target agent loops directly: the database auto-detects agent harnesses, returns structured errors with explicit retryable flags, and enforces a row cap with an in-band truncation marker so large retrievals do not fill a context window. A single Cypher statement can combine vector similarity search with graph traversal, and durability is optional through an fsync’d write-ahead log or a flush to disk. Neo4j states that Relate is 100% openCypher and more than 97% GQL compliant, with sub-millisecond in-memory reads; the press release gives no benchmark methodology for the latency figure. Press release Repository
2. Pomerado Launches an MIT-Licensed Harness That Turns a Website Into a Deterministic MCP Server
Pomerado (YC F26) launched on Hacker News on November 11 with an open-source core (Pomerado/pomerado, MIT) and a free hosted version. A user or a coding agent describes a task on a website in natural language; a Pomerado agent performs it once in a browser by writing HTTP requests or a Playwright script, then generalizes the recorded run into a repeatable API exposed over MCP, a process the founders say takes under an hour. Later tool calls run the deterministic integration without a model in the loop and, per the launch post, usually return in 5 to 30 seconds. A separate “Guardian” model reviews the building agent’s calls against guardrails, synthetic-input tests check that the integration is not overfit to one input, and sign-in is handed to a deterministic process so the model and recorders do not read credentials; users can store TOTP seeds for two-factor prompts. When a later run fails, a maintenance job repairs the integration. The speed and cost comparisons against computer-use agents on the project site are the company’s own and have not been independently reproduced. Repository Launch HN
3. Talos 0.20.0 Becomes the First Stable Release of a Self-Hosted Agent Gated by a Deterministic Capability Kernel
Talos (talos-kernel/Talos, MIT) published version 0.20.0 on November 11, which the maintainers describe as the first stable release under a frozen operator contract. Talos is a self-hosted agent for the terminal, Telegram and Raspberry Pi that runs local Ollama models or Claude and Codex workers, with every tool call passing through a deterministic capability kernel, task-scoped approvals and an audit trail. The stable release carries forward the release-candidate runtime: migration from Hermes and OpenClaw setups, Talos Connections, managed skills, labelled scheduled results and a pypdf security fix. The release notes list the qualification behind the “stable” label as seven daily deployment records from November 5 to 11, full Python 3.12 and 3.13 and ARM64 Linux suites, 277 adversarial cases, 44 isolated real-model end-to-end checks, and signed clean-install, upgrade and rollback rehearsals. Release assets are a signed archive with a SHA-256 checksum and an Ed25519 signature; experimental computer adapters keep separate qualification requirements. Release