vectorize-io/hindsight#coding-agents
Hindsight, agent memory that learns: long-term project memory with auto recall and retain, knowledge pages, deep reflection, and per-repo memory banks.
Install
via GitHub · GitHub source
This repository ships several plugins; the suffix selects this one.
Due diligence
coding-agents is a DeepSeek Harness plugin that adds Hindsight’s long-term memory system to coding agents, with recall, retention, knowledge pages, reflection, and separate memory banks for repositories.
Hindsight can sit behind an agent through an LLM wrapper, which automatically stores and retrieves memories as the agent makes LLM calls. For more control, its clients expose retain, recall, and reflect, with a bank_id selecting the memory bank. The README documents a local Docker service at ports 8888 (API) and 9999 (UI), Python and Node.js clients, and an embedded Python mode using HindsightServer. The documented provider configuration includes HINDSIGHT_API_LLM_PROVIDER; API access requires a provider credential such as OPENAI_API_KEY or HINDSIGHT_API_LLM_API_KEY.
This is intended for agents that need project context across interactions, user or task memories, and behaviour that can change from accumulated information. It is a poor fit for a small workflow that only needs simple retrieval, since the README describes Hindsight as potentially overkill for such applications. The terminal surface is used to run its Docker, Python, and npm tooling, and the service may send memory-processing requests to the configured LLM provider, so credentials and the agent’s terminal permissions should be treated as sensitive. coding-agents also introduces a separate memory service rather than using only the harness’s existing session state.
Hindsight stores information with retain, searches stored memories with recall, and generates a disposition-aware response with reflect. Each operation takes a bank_id; the README uses my-bank. Memory banks can therefore be selected per repository or other scope by the calling agent. An LLM wrapper is also available: it replaces the existing LLM client and stores and retrieves memories automatically during LLM calls.
The Docker quick start exposes the following settings:
| Setting | Meaning |
|---|---|
OPENAI_API_KEY |
Shell variable used in the example as the provider credential |
HINDSIGHT_API_LLM_API_KEY |
API container variable receiving the LLM API key |
HINDSIGHT_API_LLM_PROVIDER |
Selects the LLM provider |
The README lists these provider values for HINDSIGHT_API_LLM_PROVIDER: openai, anthropic, gemini, groq, ollama, lmstudio, minimax, and atlas.
The Docker deployment serves the API at http://localhost:8888 and the UI at http://localhost:9999. The recommended container mounts the named volume hindsight-data at /home/hindsight/.pg0. An external PostgreSQL deployment is documented through docker/docker-compose; Oracle AI Database is also listed as supported for enterprise deployments.
Python clients use Hindsight from hindsight_client, while Node.js clients use HindsightClient from @vectorize-io/hindsight-client. The README also documents embedded Python operation through HindsightServer, which starts a local server for the client to use.
The embedded package is hindsight-all. On Intel (x86_64) Macs, the README directs users to hindsight-all-slim instead. The examples use Python packages hindsight-client and hindsight-all, or the npm package @vectorize-io/hindsight-client.
Written from the project's own documentation and kept in sync with it. Where the two disagree, the source is authoritative — read the README on GitHub
Same category
volcengine
OpenViking memory and context bundle for DeepSeek Harness: pre-step auto-recall and profile injection, session capture, `viking://` URI guarding, and recall/write memory tools backed by an OpenViking server.
tinqiao-oss
The Engramory curated-memory discipline as an installable plugin ([npm: dsh-engramory](https://www.npmjs.com/package/dsh-engramory)): a deterministic 200-line / 25 KB cap on the `MEMORY.md` index via `ctx.tools.guard()` — growth denied, a shrinking rewrite always passes — plus the protocol registered as a runtime skill. The store is plain markdown, one file per fact, shared with Claude Code, Codex, Kiro, and OpenClaw.
bowenliang123
Context insight panel: see what the model's context window is made of and how it evolves — composition vs. window size, per-request history, compression/injection events, and per-message token stats.
Co-Engram
Self-evolving team memory as plain Markdown in git: a native Cordis plugin that registers 38 bare-name memory tools plus a prompt-signals section re-evaluated at every assembly, with RPE reinforcement, decay and sleep consolidation; shares one data repo with its Claude Code (MCP) and OpenClaw hosts; verified against DSH 0.1.0-rc.6.
omdsh-dev
Cross-agent, local-first persistent memory plugin for DeepSeek Harness (DSH), powered by Mnemon. It shares long-term memory across Mnemon-enabled agents and adds runtime memory, searchable project documents, semantic recall, knowledge graph, and a Sidebar UI.
Aik358
Auto-memory for DSH: three-layer memory (user / project notes / daily logs) with automatic injection, per-turn auto-consolidation, AI greetings, smart search, a calendar view and a settings page, plus inheritance of other AI tools' memories.
