Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Get Started Free →

Add an AI agent / RAG backend · get-convex | DecimalAI

get-convex/convex-agent

SkillScore91Maintained

Add an AI agent / RAG backend (@convex-dev/agent) to the Convex app.

agent-toolingby get-convex

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

AI Engineer pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/get-convex-convex-agent/SKILL.md

decimalai skills pull get-convex-convex-agent --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/get-convex-convex-agent/SKILL.md https://app.decimal.ai/s/get-convex-convex-agent/SKILL.md

Raw file: /s/get-convex-convex-agent/SKILL.md ↗ · machine-readable summary at /s/get-convex-convex-agent.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 22d ago

Yes — +50% lift-23% tokens0% turns91% of whole cases pass with skill (20/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-01	✗→✓	▲ Improved	-34%	0%
case-02	✗→✓	▲ Improved	-1%	0%
case-03	✗→✓	▲ Improved	-13%	0%
case-05	✗→✓	▲ Improved	-61%	0%
case-06	✗→✓	▲ Improved	-48%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v5 — 8/30/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

get-convex / agent-skills

skills/convex-agent/SKILL.md

View on GitHub ↗

Imported 8/30/2026Version v2

Skill contents

SKILL.md

Add an AI agent / RAG backend

Install @convex-dev/agent for durable threads, message history, tool-calls, and vector search/RAG — the backend for an in-app AI agent. Call models through the Convex AI Gateway by default: Convex holds the provider credentials, so there is no LLM key to obtain, store, or rotate.

Workflow

Install @convex-dev/agent + @convex-dev/ai-sdk-provider; add the agent component to convex.config.ts.
Define the agent (tools, instructions) with languageModel: convexGateway("provider/model") — no API key needed (needs convex 1.45+ on a Convex Cloud deployment, paid plan).
Create threads + stream messages; persist history in Convex.
For RAG: embed docs into a vector index and retrieve in the tool. The gateway does not serve embeddings yet, so store the embedding provider's key via the env micro power.
Only if the gateway is unavailable (free plan, self-hosted, local backend): call the provider SDK with a key stored via the env micro power.

Rules

Default to the Convex AI Gateway (convexGateway from @convex-dev/ai-sdk-provider) for model calls; fall back to a provider key in Convex env only where the gateway is unavailable (free plan, self-hosted, local backend).
Never expose a provider API key client-side; when one is needed (embeddings, gateway fallback), keep it in Convex env via the env micro power.
Run model calls in actions ('use node' if the SDK needs it).
Persist threads/messages in Convex for durability + reactivity.

Illustrative Want to see what get-convex/convex-agent would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/30/2026, 06:10 UTC · 22 cases

Lift

+50%

91% vs 41%

Agent turns

0%

1.0 vs 1.0

Tokens

-23%

1,850.32 vs 2,408.68

fail → pass · 13pass → pass · 7pass → fail · 2

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 I'm building an in-app AI assistant using Convex and need to set up durable thread history and model integration. Can you provide a code example and guide on how to configure the backend agent and run model calls securely without exposing API keys to the frontend?	fail→pass	24,087	15,837	-34%	1	1	0%	4,157	2,728	-34%	0	0	—
case-02 We want to add vector search and document retrieval (RAG) to our Convex backend AI agent. How should we set up the agent configuration, store embedding credentials safely, and connect vector retrieval to tool calls?	fail→pass	22,076	19,942	-10%	1	1	0%	3,538	3,496	-1%	0	0	—
case-03 Can you show me the step-by-step backend setup for an AI chat agent in Convex? I need to know how to initialize the agent component in my config, define instructions and tools, and handle thread streaming and persistence.	fail→pass	20,437	16,026	-22%	1	1	0%	3,124	2,718	-13%	0	0	—
case-04 I am configuring a Convex backend project to support AI chat capabilities. How do I enable the backend agent capabilities in the backend configuration file?	pass→pass	13,248	8,751	-34%	1	1	0%	1,608	1,141	-29%	0	0	—
case-05 We are deploying an AI chat feature on a Convex Cloud paid project. We want to avoid managing, storing, and rotating direct OpenAI or Anthropic API keys. Which provider wrapper should we pass as the languageModel parameter when instantiating the agent?	fail→pass	19,590	8,378	-57%	1	1	0%	2,685	1,047	-61%	0	0	—
case-06 Our team wants to use the zero-API-key AI gateway feature for model calls in our Convex backend. What are the deployment plan and framework version requirements for this feature to work?	fail→pass	16,408	8,210	-50%	1	1	0%	1,880	977	-48%	0	0	—
case-07 We are running local development testing for our Convex backend where the AI Gateway cannot be reached. How should model credentials and provider instantiation be handled in this local environment?	pass→fail	20,750	14,979	-28%	1	1	0%	2,936	2,473	-16%	0	0	—
case-08 We need to generate document embeddings for vector search in our Convex AI agent. Can the default Convex AI Gateway serve embedding calls directly, or how should embedding credentials be configured?	pass→pass	17,651	9,533	-46%	1	1	0%	2,416	1,287	-47%	0	0	—
case-09 When writing the Convex backend functions that invoke language model calls, what type of Convex backend function should execute these model calls, and what directive might be needed if the underlying SDK depends on Node.js APIs?	pass→pass	11,282	8,933	-21%	1	1	0%	1,185	1,034	-13%	0	0	—
case-10 We need reactive message history and durable thread storage in our Convex backend for an interactive assistant. Should we write custom schema tables and indexes in schema.ts, or is there a built-in package solution?	fail→pass	17,884	10,515	-41%	1	1	0%	2,251	1,499	-33%	0	0	—
case-11 To reduce backend latency, a developer suggests putting OpenAI API keys into frontend environment variables so the browser can stream completions directly. How should API keys be handled in this architecture?	fail→pass	16,005	12,343	-23%	1	1	0%	1,975	1,762	-11%	0	0	—
case-12 When setting languageModel on a Convex AI agent using the built-in gateway, what is the exact string format passed to the gateway function to specify a model like GPT-4o or Claude 3.5 Sonnet?	fail→pass	9,236	3,979	-57%	1	1	0%	1,826	1,208	-34%	0	0	—
case-13 How should document vector retrieval be hooked into an agent created with Convex agent utilities so the model can retrieve relevant context dynamically?	pass→pass	20,408	12,684	-38%	1	1	0%	3,133	3,001	-4%	0	0	—
case-14 What primary npm packages must be installed in a Convex project to set up durable agent threads and leverage the zero-config AI gateway?	fail→pass	11,710	8,641	-26%	1	1	0%	1,162	1,072	-8%	0	0	—
case-15 Our project is running on a free Convex Cloud plan. Can we use the zero-API-key AI Gateway without providing an API key, or what is the correct fallback setup?	fail→pass	16,836	7,465	-56%	1	1	0%	2,015	1,986	-1%	0	0	—
case-16 Where are system instructions and custom executable tools defined when configuring an AI agent in a Convex backend?	fail→pass	18,276	7,053	-61%	1	1	0%	2,416	1,746	-28%	0	0	—
case-17 How does the model provider configuration differ when executing an AI agent on a paid Convex Cloud instance versus a local offline Convex backend?	fail→pass	21,020	8,954	-57%	1	1	0%	3,128	2,145	-31%	0	0	—
case-18 How does the Convex agent architecture handle thread persistence and UI synchronization across client sessions?	pass→pass	16,864	16,198	-4%	1	1	0%	2,816	2,305	-18%	0	0	—
case-19 When implementing a Convex action that uses Node-dependent AI provider SDKs for local development fallbacks, what directive must be included at the top of the file?	pass→fail	8,458	7,371	-13%	1	1	0%	621	779	+25%	0	0	—
case-20 We want to use the zero-API-key Convex AI Gateway feature to generate text embeddings for documents before inserting them into our vector database. How should we configure the gateway for embedding generation?	fail→pass	24,626	12,912	-48%	1	1	0%	3,242	2,052	-37%	0	0	—
case-21 We need a Convex backend action that generates a one-off text summary of a document string on demand. The feature does not require message history, thread persistence, reactive UI streaming, or tool execution. Should we configure a persistent agent component with thread storage for this stateless task?	pass→pass	13,459	12,950	-4%	1	1	0%	1,538	1,632	+6%	0	0	—
case-22 We have a vector index 'by_embedding' defined on a Convex 'documents' table. How do we execute a vector search query function in Convex to retrieve matching documents without using an AI agent component or agent tool wrappers?	pass→pass	21,523	15,729	-27%	1	1	0%	3,339	2,619	-22%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +50 percentage points is the difference between those two pass rates over the 22 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 40.9%
Pass rate with the skill: 90.9%
Cases attempted: 22
Cases counted toward lift: 22
Cases that regressed: 2
Skill version measured: v2
Run completed: 8/30/2026, 06:10 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-01

Prompt

I'm building an in-app AI assistant using Convex and need to set up durable thread history and model integration. Can you provide a code example and guide on how to configure the backend agent and run model calls securely without exposing API keys to the frontend?

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run91

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns-19% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 22 days ago

Model	Method	Date	Lift
gemini-3.6-flash	verified	8/28/2026	+17%

Other measured skills in the registry, with their headline benchmark lift.