Veecle
Also known as: Veecle GmbH, Chiplab
Agent infrastructure for firmware: Chiplab, an MCP server that lets AI coding agents run and test firmware on virtual microcontrollers without physical hardware.
Veecle builds Chiplab, a hosted MCP server that lets AI coding agents run firmware on virtual instances of real microcontrollers instead of physical boards. An agent connected through Cursor, Claude Code, Codex, VS Code, OpenCode or any other MCP capable client builds an ELF, uploads it and runs it on a chip accurate virtual board (STM32 and Nordic nRF52 today, simulated on Renode), then reads back the captured UART output and simulator diagnostics.
A second tool, ask, answers questions about the platform, supported boards and firmware patterns from Chiplab's documentation and a shared corpus of chip level behavior observed across runs. Veecle says the corpus never collects customer firmware or binaries. Agents sign in through a browser based OAuth flow, and the dashboard lists connected agent sessions with revoke and shows request history for the billing period. Chiplab is in beta.
Remote builds and tests are in progress, board benchmarking and full system simulation with partner environment models are planned next, and an orchestration tool is on the roadmap from 2027. A public GitHub repository carries ready to run examples for every supported board across bare metal and Embassy Rust, Zephyr, FreeRTOS and ThreadX. Pricing is credits based with a free daily pool. Paid top ups are announced as coming soon, and an Enterprise tier offers on premises deployment. Veecle is a Berlin company whose earlier work, an open source Rust RTOS and a browser based embedded IDE, fed into Chiplab.
Vendor details
Canonical URL
https://veecle.ai
Category
Agent infrastructure
Funding status
Independent. Veecle GmbH is based in Berlin, and its about page names backing from exist, Relay Ventures, IBB Ventures, F-LOG Ventures, Plug and Play and IT Inkubator. No round sizes are stated.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Chiplab is reached only through its own HTTP MCP server (chiplab.veecle.ai/mcp) with browser based OAuth sign in, with setup guides for Cursor, Claude Code, Claude Desktop, Codex, VS Code and OpenCode and support for any MCP client. Simulation runs on Renode today. There is no connector catalog. Partner environment models (BeamNG, dSPACE, NVIDIA Isaac) are on the roadmap for a planned simulate tool.
In practice
Your coding agent changes firmware but has no board to try it on. Chiplab lets the agent build an ELF, upload it and run it on a chip accurate virtual board, then read back the console output and fault diagnostics.
Your team wants every firmware change tested before the agent reports back. Chiplab's setup guides add a rule to CLAUDE.md, AGENTS.md or a Cursor rule so the agent runs the change on a virtual board first.
Your agent hits a fault it does not recognize. Chiplab's ask tool answers from its documentation and a shared corpus of chip level behavior observed across runs, without collecting your firmware or binaries.
Sources & related URLs
Related / legacy domains
Agentic Index coverage score
3.0 / 14 capabilities · 21%
| Integrations & Tool Calling | Not documented |
|---|---|
|
Chiplab is itself the tool an agent calls. Its MCP server exposes ask, a knowledge base assistant, and run, which runs firmware on a virtual board. The customer's agent host decides what else the agent reaches. Chiplab has no connector catalog, custom tool route or action framework of its own for taking authenticated action in outside systems. Renode, the simulator behind run, is Chiplab's execution backend, not a system the agent acts in. The Enterprise tier's custom chip and tool integrations add targets to Chiplab and do not connect the agent to other systems. Partner environment models are planned for a future simulate tool. Sourcedocs.veecle.ai/tools/overviewread 2026-09-23 |
|
| Workflow Orchestration | Not documented |
|
Chiplab does not run multi step work of its own today. A run may be a single call or a short sequence (issue an upload slot, upload the artifact, trigger the run) that the customer's agent discovers and sequences itself. An orchestrator tool called automate, which would chain build, run, test, bench, simulate and ask, is on the roadmap. Sourceveecle.ai/roadmapread 2026-09-23 |
|
| Knowledge Grounding & RAG | Not documented |
|
The ask tool routes a question to an assistant scoped to Chiplab usage topics. It draws on Chiplab's documentation and a shared knowledge corpus of chip level behavior observed across the platform's runs, indexed by chip family, failure pattern and board configuration. The corpus is pooled across every user and, by Chiplab's own terms, never collects the customer's firmware or binaries. Veecle names no retrieval structure over the customer's own corpus. Sourcedocs.veecle.ai/tools/askread 2026-09-23 |
|
| Human Oversight & Guardrails | Not documented |
|
The agent uploads, runs and reads results on its own, and the customer does not need to call anything themselves. Veecle names no step where a person reviews, approves or pauses a run. The dashboard can revoke a connected agent session, which controls access but does not review the work. Sourcedocs.veecle.ai/tools/runread 2026-09-23 |
|
| Security, Identity & Governance | Not documented |
|
Agents authenticate to Veecle's own accounts through a browser based OAuth 2.0 flow. Chiplab offers no sign in through the customer's own identity provider and no named access model with roles or permissions. The API Keys page lists authorized agent sessions with first connected and last active times and a revoke control. It manages credentials on a single account and does not show who in the customer's organization can do what. Veecle names no certification or compliance attestation, shows no certification badge or trust link and publishes no trust center. Its privacy notice is a GDPR controller statement that names processors (Supabase in Frankfurt or Ireland, Cloudflare, Stripe) and a retention schedule. Sourcedocs.veecle.ai/platform/api-keysread 2026-09-23 |
|
| Observability & Auditability | Partial |
|
The agent's requests are recorded. The Usage dashboard loads the account's full request history for the current billing period, with aggregate totals and a recent window for confirming a run went through. The API Keys page shows each agent session's first connected and last active times. Together they record what the agent did on Chiplab over a billing period. The agent's own host handles step by step inspection of prompts, tool calls and outputs. Veecle names no audit log separate from usage, no export to a monitoring system and no retention by plan. The console output and fault diagnostics a run returns describe the firmware's behavior, not the agent's. Sourcedocs.veecle.ai/platform/usageread 2026-09-23 |
|
| Memory & State Persistence | Not documented |
|
Runs return synchronously, bounded to a fixed amount of virtual CPU time. Veecle names no state carried from one run into the next. Each run also feeds the shared knowledge corpus, which is learning pooled across the platform, not a memory the customer's agent reads, reviews or edits. Chiplab describes each run as an addressable object with an identity of its own, but names no way to resume a held chip from that handle. Sourcedocs.veecle.ai/introductionread 2026-09-23 |
|
| Deployment & Data Residency | Full |
|
Chiplab is a hosted MCP service. Its Enterprise tier offers on premises deployment alongside volume commits, custom chip and tool integrations and SLAs, so customers can run it in their own environment. Veecle does not say how an on premises install is delivered, and no deployment guide is published. The hosted service's database and authentication sit with Supabase in the EU (Frankfurt or Ireland). That location is part of Chiplab's own infrastructure, and customers do not select it. Sourceveecle.ai/pricingread 2026-09-23 |
|
| Prebuilt Agents / Templates / Packs | Partial |
|
A public examples repository on GitHub carries a ready to run firmware example for every supported board across five frameworks (bare metal and Embassy Rust, Zephyr, FreeRTOS and ThreadX). Each comes with an AGENTS.md that tells a coding agent how to build, upload and run it. A developer's agent builds from these examples, and there is no catalog of agents to adopt. Sourcegithub.com/veecle/chiplabread 2026-09-23 |
|
| Triggers & Channel Coverage | Not documented |
|
Work starts only when the customer's agent calls. The coding agent discovers the tools and calls them when a message involves firmware simulation. Chiplab has no event, schedule, webhook or inbound queue of its own that starts work. Veecle suggests adding a rule to CLAUDE.md, AGENTS.md or a Cursor rule that tells the agent to test changes on Chiplab before reporting back, and that rule lives in the customer's agent. Per commit testing is planned for the test tool, which the roadmap lists as in progress. Sourcedocs.veecle.ai/agents/overviewread 2026-09-23 |
|
| Model Flexibility & Routing | Not documented |
|
Any MCP capable coding agent can call Chiplab, including Cursor, Claude Code, Codex, VS Code and OpenCode, so the customer picks the assistant and its model. The assistant behind ask names no model and offers no choice of one. Sourcedocs.veecle.ai/agents/overviewread 2026-09-23 |
|
| APIs / SDKs / MCP Extensibility | Partial |
|
Outside access runs only through Chiplab's own HTTP MCP server at chiplab.veecle.ai/mcp. Agents sign in through a browser based OAuth 2.0 flow with automatic client registration, and setup guides cover six clients. The two tools, ask and run, are listed with their parameters and response fields. Veecle calls those tool names, parameters and response shapes illustrative, not a frozen API contract. The agent makes the calls, and customers make no manual API calls. Veecle names no REST API or SDK, and the API Keys page manages OAuth agent sessions, not API keys. What the home page calls a single API is the MCP endpoint. Sourcedocs.veecle.ai/tools/overviewread 2026-09-23 |
|
| Testing, Debugging & Optimization | Partial |
|
The run tool tests the customer's own firmware. It executes what the agent built on a chip accurate virtual board, with the same binary, peripherals and interrupt timing as the physical part, and returns console output with simulator diagnostics for faults. Every agent setup guide suggests gating changes on it, telling the agent to test firmware changes on Chiplab before reporting back. A test tool for unit and integration tests, coverage and regression tracking is on the roadmap as in progress. The run tool returns a run ID, stdout and stderr, with no pass or fail status. Today a run gives console output and fault diagnostics to read afterwards. Sourcedocs.veecle.ai/tools/runread 2026-09-23 |
|
| Browser / Computer-use | Not documented |
|
Agents reach Chiplab through MCP tool calls that upload an ELF and run it on a virtual board. The virtual chip is an execution target driven through those calls, not a screen the agent navigates. Veecle names no browser or desktop the agent operates, and no screenshot, click or typing surface. Sourcedocs.veecle.ai/introductionread 2026-09-23 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Product updates
Veecle's Chiplab added a run_test_suite MCP tool that lets a coding agent write a TypeScript test suite, run it against uploaded firmware on a virtual board, and get back a pass or fail for each test along with the serial output. Each suite has a five minute time limit.
Bears on: Agent capability
View sourcePricing
Free: €0 forever with 1,000 credits a day; paid top-ups (€10 per 1,000 credits) announced as coming soon
Credits drawn per MCP tool call, priced by the compute each call triggers
Cost watchouts
Every tool call draws credits priced by the compute it triggers, and the per-call credit cost is not published, so an agent told to test every firmware change on Chiplab spends in proportion to its loop. Paid top-ups are not live yet, so today's ceiling is the free daily pool.
Variable cost rationale
Spend scales with how often the customer's agent calls Chiplab, since each call draws credits by the compute it triggers; the free daily pool caps spend at zero today because paid top-ups are announced but not live.
Sales call required
No, self serve available
Free / trial
Free tier: €0 forever, 1,000 credits a day, self-serve sign-up with no credit card
Key ambiguities
The credit cost of each tool call is not published, so how many runs 1,000 daily credits buy is unknown; the Pay-as-you-do price is announced as coming soon, not a live plan.
Related vendors
- AgentOps — Agent observability and debugging platform: open source SDKs trace…
- AgentPhone — AgentPhone gives the AI agents a team builds their own US or…
- Agno — Python agent framework and AgentOS runtime (formerly Phidata) for…
- AIsa — Resource and payment gateway for AI agents, with one key to 110+…
- Albato — Albato is a no code integration platform that sells an embedded,…
- AlphaBitCore — AI control plane for regulated financial firms: one gateway enforces…
Alternatives to Veecle
The closest documented capability profiles to Veecle among agent infrastructure platforms tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Corellium4.5 / 14Adds documented Security, Identity & Governance and Memory & State PersistenceVeecle vs Corellium →
- Rime3.5 / 14Adds documented Security, Identity & Governance
- Speechmatics3.5 / 14Adds documented Security, Identity & Governance
- Wokwi3.5 / 14Adds documented Memory & State PersistenceVeecle vs Wokwi →
- AgentOps5.0 / 14Adds documented Integrations & Tool Calling and Security, Identity & Governance, among others
- Patronus AI0.0 / 14A lighter documented profile than Veecle
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded