Every HeliosDB product ships an AGENTS.md aggregate plus a .claude/skills/ catalogue. Every CLI flag, REPL meta-command, public API, and MCP tool is indexed for LLM coding agents — no README spelunking required.
Each card shows the install path for that product. HeliosProxy and Lite unpack a skill bundle embedded in their binary, Nano installs its skills from a checkout of its public repository, and Full ships an installer script with its licensed distribution.
Embedded multi-model database. Skills cover overview, install, connect, schema, query, transactions, branches, time-travel, backup, vector, code-graph, graph-rag, mcp, server, deploy, observability, migrate, merge-validation, tenant. The script copies them into ~/.claude/skills/ (add --symlink to link instead); the heliosdb-nano binary has no skills subcommand.
git clone https://github.com/HeliosDatabase/HeliosDB-Nano
bash HeliosDB-Nano/scripts/install-agent-skills.shProduction self-hosted. Skills span foundation, DML/DDL, ops, advanced features, security, HA/replication, AI/agent surface — in seven phases. The heliosdb-lite binary is provided to licensed customers through sales (it is not on crates.io); the install subcommand takes the destination directory, and --catalog codex selects the Codex catalogue.
# licensed binary from HeliosDB sales
heliosdb-lite skills install ~/.claude/skillsEnterprise edition. The catalogue mirrors Lite and adds client skills for Full's listeners. HeliosDB Full serves PostgreSQL, MySQL, Oracle and Redis clients plus an HTTP gateway; MongoDB, Cassandra and SQL Server compatibility is planned. Full is provided to licensed customers during onboarding (it is not on crates.io), and its installer script links the skills into ~/.claude/skills/heliosdb-full/.
# in your licensed Full distribution:
# run the bundled skills installer scriptPostgres operator playbook. Pool tuning, failover rehearsal, time-travel replay, plugin signing, OCI distribution, PG-12→17 upgrade orchestration.
cargo install heliosdb-proxy
heliosdb-proxy install skillsSKILL.md per operational domainA skill is one Markdown file scoped to one verb or workflow. The agent reads it once and knows the exact API: which command to run, which flags matter, which failure modes to look for.
# heliosdb-nano-vector / SKILL.md (abridged)
## When to use
- Similarity search over text/image/audio embeddings.
- Building the retrieval half of a RAG pipeline.
## Verbs
| create HNSW index | SQL | CREATE INDEX vidx ON t USING HNSW (col)
| | | WITH (dim = 384, metric = 'cosine')
| distances | SQL | ORDER BY embedding <-> / <#> / <=> $1 LIMIT k
| insert vectors | Rust | db.insert_vectors("docs", vec)?
| hybrid search | MCP tool | heliosdb_hybrid_search (mcp-endpoint feature)
## Recipe: top-k similarity query (cosine)
SELECT id, title, body
FROM docs
ORDER BY embedding <=> $1 -- $1 = query embedding
LIMIT 5;
## Pitfalls
- Dimension mismatch fails at insert time, not at index creation.
- <-> is L2; <#> is negated inner product (smaller = more similar).
- HNSW indexes are not the right tool for tiny tables (a few thousand vectors).… mediocre at exploring unfamiliar APIs. Without a skill catalogue, an agent fumbles through HeliosDB the same way a new hire does. With it, the agent reads AGENTS.md and knows the verb for the task in front of it.
Each skill names a single verb — BRANCH, WITH CONTEXT, graph_bfs. No "see also" trees. Find the verb, get the API.
Every skill ends with the failure modes most likely to confuse an agent — dimension mismatches, FK in transactions, branch-aware visibility.
_index/verb-map.md normalises every CLI flag ↔ SQL function ↔ MCP tool. _index/feature-matrix.md maps cargo features to skills.
If a feature ships without a skill, it doesn't ship. Skill counts grow with every release — treat them as part of the API.
These flags belong to heliosdb-proxy install skills. Nano's script takes only --symlink and backs up existing heliosdb-nano-* skills to *.bak.<ts>; Lite's skills install <target> takes --symlink, --catalog claude|codex and --force.
| Flag | Effect |
|---|---|
--target claude|codex|both | Default both. Where the bundle lands. |
--symlink | Symlink instead of copy. Re-running after a binary upgrade refreshes in place. |
--force | Overwrite existing heliosproxy-* skills at the target. Without it, existing entries are skipped and reported. |
--dry-run | Print the planned actions without writing. |
Skill bundles teach an agent how to use HeliosDB. HeliosDB CodeKB MCP turns the agent's loop inside-out: it indexes your repo into a local KB and exposes 11 LSP-shaped + Graph-RAG tools so Claude Code, Codex, Cursor, and Aider can reason about the code you're actually editing.
curl -L https://github.com/HeliosDatabase/HeliosDB-CodeKB-MCP/releases/latest/download/heliosdb-codekb-mcp-linux-x86_64 \
-o /usr/local/bin/heliosdb-codekb-mcp && chmod +x /usr/local/bin/heliosdb-codekb-mcp
heliosdb-codekb-mcp init --source ~/code/myrepo --mode co-located --ingestApache 2.0 · embeds HeliosDB-Nano · Full page →
Re-run your product's skills installer after an upgrade and your agent stack catches up to the latest verb catalogue.