Foundry is live · custom pricing · last verified 23 Oct 2026 · viability monitor
Foundry
Browser-agent simulation and evaluation on pixel-perfect, reproducible enterprise SaaS environments.
If your agent team is burning weeks on flaky browser tests, Foundry's resettable environments and per-action classification target exactly that pain, and the evaluation_loop.py pattern plugs into an existing Python agent loop without a rewrite. SDRBench adds a shared, deterministic measuring stick across Gmail, Sheets, Salesforce, and Apollo. The trade-off versus WebArena and BrowserGym is fidelity on named enterprise platforms rather than a free, widely cited public suite.
Verdict generatedVerified 7d ago · liveness 63/100 · cite: rightaichoice.com/tools/foundry
- Agent teams whose evals break because live SaaS environments drift mid-run
- Enterprise developers shipping browser agents into Salesforce, Gmail, or Apollo workflows
- RL researchers who need unlimited trajectory sampling without anti-bot blocking
- Teams that want expert-annotated training data on real enterprise platforms
- Teams without Python engineers — the SDK and evaluation loop are code-first
- Anyone whose agents never touch a browser
- Academic researchers who primarily need a free, widely cited public benchmark
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Foundry if your evals are already stable and your agents rarely touch third-party SaaS, or if you have no Python engineer to wire the SDK's evaluation loop into your stack.
Wiring the Python SDK and evaluation_loop.py into your stack is engineering time, not a plug-in install, so budget a developer before you start.
Foundry does not publish tier or billing structure on the pages we could reach, and access runs through a private beta application. That makes it a fit for funded agent teams and enterprise platform groups who can absorb a procurement and integration cycle, rather than for solo builders comparing it against a free public benchmark suite.
In short
Foundry — Browser-agent simulation and evaluation on pixel-perfect, reproducible enterprise SaaS environments. Best for Agent teams whose evals break because live SaaS environments drift mid-run, Enterprise developers shipping browser agents into Salesforce, Gmail, or Apollo workflows, RL researchers who need unlimited trajectory sampling without anti-bot blocking. Pricing is on request (custom quotes).
What is Foundry?
Foundry — Browser-agent simulation and evaluation on pixel-perfect, reproducible enterprise SaaS environments. Best for Agent teams whose evals break because live SaaS environments drift mid-run, Enterprise developers shipping browser agents into Salesforce, Gmail, or Apollo workflows, RL researchers who need unlimited trajectory sampling without anti-bot blocking. Pricing is on request (custom quotes).
Foundry is a simulation and data engine for teams building AI web agents that operate real enterprise software. Rather than pointing agents at live SaaS accounts and hoping nothing shifts mid-run, you run them against pixel-perfect, reproducible browser environments with no drift, no noise, and no rate limits. Four pillars carry the product: simulation, evaluation, data, and RL. Simulation hands you resettable browser states at production fidelity. Evaluation tracks, classifies, and tags every agent action, so click failures, layout shifts, and misfires surface instead of disappearing into a log. The data layer pairs expert annotators with real enterprise platforms to produce long-horizon datasets for supervised fine-tuning. The RL layer lets you sample and evaluate unlimited trajectories at scale without anti-bot constraints. Integration is a Python SDK that drops into existing agent loops. The published evaluation_loop.py pattern wires an env object with a task_id, pulls task.prompt, task.start_url, task.login_credentials, task.payment_details, and task.additional_instructions, then compares the final state against task.ground_truth with a heuristic score. Foundry also publishes SDRBench, a benchmark and RL gym for browser agents modeling enterprise SaaS workflows across Gmail, Sheets, Salesforce, and Apollo, with 50 deterministic tasks and leaderboard baselines. It is built for agent developers and platform teams needing consistent results across hundreds of runs, and for researchers wanting a stable environment to compare approaches. The company is YC-backed and staffed from the browser-agent research world. Academic suites like WebArena and BrowserGym optimize for standardized tasks in controlled settings; Foundry leans toward production fidelity on named enterprise platforms, with simulated login credentials and payment flows baked into the task structure.
Key features
- Pixel-perfect reproducible browser environments for agent simulation
- Resettable environment state between evaluation runs
- No drift, noise, or rate limits across runs
- Per-action tracking, classification, and tagging of agent behavior
- Click failure, layout shift, and misfire detection
- Expert-annotated long-horizon datasets for supervised fine-tuning
- Datasets built on real enterprise platforms
- Unlimited trajectory sampling and evaluation for RL
- RL gym without anti-bot constraints
- Python SDK that plugs into existing agent workflows
- CDP URL retrieval to control and observe browser agents
- Task structure carrying login credentials and payment details
- Ground-truth comparison via heuristic scoring
- Event recording with append-only state mutation history
- SDRBench benchmark and RL gym with 50 deterministic enterprise SaaS tasks
Behind the verdict
Reach for Foundry when your browser-agent evals keep breaking because the underlying SaaS environment moved under you. The pitch is narrow and honest: reproducible state, per-action classification, and unlimited trajectories without anti-bot walls. For a team that has been re-running the same Salesforce flow and getting different numbers each time, that's a real fix, not a positioning line. We'd pick it when the work is code-first and browser-heavy. The evaluation_loop.py pattern is the tell — env objects, task_id, get_cdp_url, get_state, heuristic scoring against ground truth. If your agents live in a Python loop and touch real web apps, this slots in fast. The 5-minute setup claim is plausible for that audience. Where it bites: there's no no-code path, and the SDK assumes Python engineers on staff. Teams whose agents never open a browser have no use for any of this. And if you mainly need a free, heavily cited public benchmark for academic comparisons, the draw here is production fidelity on commercial platforms, which is a different job. The closest alternative is WebArena or BrowserGym. Those give you a standardized, controlled task suite, often free, often cited. Foundry trades that neutrality for pixel-perfect environments that mirror Gmail, Sheets, Salesforce, and Apollo, with login credentials and payment details part of the task structure. If your production target is enterprise SaaS, that resemblance is the point. SDRBench, published in September 2025, is worth a look before you commit — 50 deterministic tasks and leaderboard baselines give you a way to see how your agent stacks up against others on the same footing. Run it, compare, then decide. On access: the platform is in private beta, and the vendor page routes new users through an application.
Limitations
- Foundry is built for one job: simulating, evaluating, and training AI web agents, so it does nothing for teams whose agents never touch a browser.
- Integration is through a Python SDK and an evaluation_loop.py pattern, which puts a coding requirement on whoever wires it up.
- The sample evaluation dataset is still labeled Coming Soon on the site, so there is no local test bundle to try before you are in.
- Access runs through a private beta application rather than an open signup.
- The reachable pages do not publish tier or billing structure, so confirm your commercial terms before you build a dependency on it.
as of 2026-10-02
Researching Foundry? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Is Foundry still active?
Foundry is live · custom pricing · last verified 23 Oct 2026 · viability monitor.
Viability score
How well maintained and how widely used is Foundry? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →What do users say about Foundry?
What people actually say about Foundry — is it worth it?
Our community sentiment research written Jul 5, 2026 is more than 90 days old, so it is being re-reviewed against current data and is not shown.
How we verified this
Every listed fact about Foundry— pricing, features, links, status — is re-read from the vendor's own pages by our verification pipeline and each change is logged with a date; the last pass ran on 23 Oct 2026. Pages mix model-assisted synthesis with human review; see how we verify.
Verification history
We have re-verified Foundry 10 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — changed description, summary, our verdict
- — changed who it suits, description, our verdict, features, hidden costs, limitations, switching in, switching out, who should skip it, pricing fit, setup time, summary, use cases
- — changed integrations, summary, who should skip it, who it suits, description, our verdict, features
- — changed SSL certificate, domain expiry
- — changed features, hidden costs, who should skip it, pricing fit, setup time, switching in, switching out, who it suits
- — changed SSL certificate
Showing the 6 most recent of 10 verification passes.
Free to cite with attribution — this page re-verifies continuously.
How much does Foundry cost in 2026?
Foundry: Pricing is on request (custom quotes). Pricing last verified 23 Oct 2026.
Where the pricing makes sense
The company stage and team size where Foundry's pricing actually pencils out — and where peers do it cheaper.
Foundry does not publish tier or billing structure on the pages we could reach, and access runs through a private beta application. That makes it a fit for funded agent teams and enterprise platform groups who can absorb a procurement and integration cycle, rather than for solo builders comparing it against a free public benchmark suite.
Has Foundry's pricing changed?
No plan price for Foundry has moved since we began logging in Aug 2026; the pricing page is re-read on every verification pass.
Who should pick Foundry?
Pick Foundry if you are Agent teams whose evals break because live SaaS environments drift mid-run, Enterprise developers shipping browser agents into Salesforce, Gmail, or Apollo workflows, RL researchers who need unlimited trajectory sampling without anti-bot blocking, Teams that want expert-annotated training data on real enterprise platforms.
Look elsewhere if you are Teams without Python engineers — the SDK and evaluation loop are code-first, Anyone whose agents never touch a browser, Academic researchers who primarily need a free, widely cited public benchmark.
Skip it if Skip Foundry if your evals are already stable and your agents rarely touch third-party SaaS, or if you have no Python engineer to wire the SDK's evaluation loop into your stack.
GeneratedReal-world workflow fit
Concrete scenarios for the personas Foundry actually fits — and what changes day-one when you adopt it.
You copy the evaluation_loop.py pattern from the site, pass your foundry_api_key and a list of task_ids into env = AWE(), call env.reset(), then pull cdp_url from env.get_cdp_url() and hand task.prompt, task.start_url, task.login_credentials and task.payment_details to your existing agent function.
Outcome: Each run returns final_state, state_mutations and events, which you score with heuristic(final_state, task.ground_truth), so a failing run comes back with a classified event list instead of a wall of log lines.
You register for the private beta, then run your model across SDRBench's 50 deterministic tasks spanning Gmail, Sheets, Salesforce and Apollo to test generalization and error recovery rather than a single happy-path demo.
Outcome: Results land on a shared leaderboard with baselines, giving you a comparable number across model versions and a reset state each time so differences reflect the agent, not the environment.
You sample unlimited trajectories from the RL layer against resettable environments, with no anti-bot blocking or rate limits interrupting long training runs.
Outcome: You can train at scale on enterprise SaaS tasks without anti-bot interference corrupting the reward signal, and inspect state_mutations when a rollout goes wrong.
Use cases
- Evaluate and fine-tune web agents for customer support interactions in a reproducible environment.
- Benchmark agent performance across realistic enterprise SaaS workflows via SDRBench.
- Generate expert-annotated long-horizon training data for custom browser agents.
- Train browser agents with reinforcement learning using unlimited simulated trajectories.
- Test agent reliability by tracking every click, layout shift, and misfire as structured events.
- Automate repetitive browser tasks such as data entry in CRM systems against simulated accounts.
- Compare two agent versions on identical reset states to isolate a regression.
- Replay a failed multi-step run using the append-only state mutation history.
Setup time & first value
How long it actually takes to get something useful out of Foundry — broken out by persona, not the marketing-page minute.
The site quotes 5min Setup Time. Realistically, a Python engineer with an existing agent loop can get a first scored run out of the published evaluation_loop.py pattern quickly; the longer path is writing or adapting task definitions and ground-truth comparisons for your own workflows. Non-engineers have no path to first value here.
Switching to or from Foundry
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From WebArena or BrowserGym: port your task definitions into Foundry's env-based task structure and point your agent at the returned cdp_url instead of a self-hosted browser.
- →From a hand-rolled Playwright/Selenium eval harness: replace the local browser setup with env.reset() and env.get_cdp_url(), and swap your own assertions for heuristic(final_state, task.ground_truth).
- →From live SaaS test accounts: move the credentials into task.login_credentials so runs execute against simulated environments rather than production tenants.
- ↗To a self-hosted browser grid: Foundry's loop already speaks CDP, so the cdp_url handoff maps to your own Chrome/Playwright infrastructure, at the cost of giving up reset states and per-action classification.
- ↗To WebArena or BrowserGym: task definitions can be re-expressed in their standardized suites if you need a free public benchmark rather than production-fidelity enterprise environments.
Integrations
Resources & guides
Tutorials & Learning
YouTube returned 6 videos for “Foundry”, and we withheld 6: 6 could not be judged, because “Foundry” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Foundry.
Official links
What are the best Foundry alternatives?
The closest alternatives to Foundry we list are Skyvern, Browserbase and browser-use. Ranked by direct product-type match, not category overlap.
| Alternative | Pricing | Starts at | Last verified |
|---|---|---|---|
| Skyvern | Freemium | Pricing not published | — |
| Browserbase | Freemium | Pricing not published | — |
| browser-use | Freemium | Pricing not published | — |
Alternatives to Foundry
View all
Skyvern
Skyvern is an open-source AI browser agent that logs into portals, solves CAPTCHAs, handles 2FA, and returns structured data.
Browserbase
Give AI agents cloud browsers to run the web like an API — logins, CAPTCHAs and forms handled.
browser-use
Browser Use gives AI agents a real anti-detect cloud browser for web automation at $0.02/hour.
Tools that pair well with Foundry
Common stack mates teams adopt alongside Foundry, with the specific reason each pairing earns its keep.
Skyvern
Skyvern is an open-source AI browser agent that logs into portals, solves CAPTCHAs, handles 2FA, and returns structured data.
Browserbase
Give AI agents cloud browsers to run the web like an API — logins, CAPTCHAs and forms handled.
browser-use
Browser Use gives AI agents a real anti-detect cloud browser for web automation at $0.02/hour.
Foundry vs its closest rival
Foundry's closest head-to-head is with Spider Cloud — Foundry is for teams building or fine-tuning browser agents with high-fidelity simulation and RL training (e.g., SDRBench for sales workflows). Spider Cloud is for AI developers who need fast, cheap web data extraction for RAG or agent context. Your choice depends on whether you need to train an agent (Foundry) or feed it data (Spider Cloud). Verdict generated
Head-to-head comparisons
Foundry vs Spider Cloud
Foundry is for teams building or fine-tuning browser agents with high-fidelity simulation and RL training (e.g., SDRBench for sales workflows). Spider Cloud is for AI developers who need fast, cheap web data extraction for RAG or agent context. Your choice depends on whether you need to train an agent (Foundry) or feed it data (Spider Cloud).
GeneratedFoundry vs Temporal Ai
Choose Temporal AI if you need a battle-tested durable execution platform for orchestrating any multi-step process—AI agents, microservices, or human-in-the-loop workflows—with automatic retries and state management. Choose Foundry if your primary need is a high-fidelity browser simulation environment for training and evaluating web agents. They serve different layers: Temporal for orchestration reliability, Foundry for agent training data and evaluation.
GeneratedFoundry vs Presto Voice
Choose Presto Voice if you run a QSR chain wanting proven drive-thru voice automation with upselling ROI. Choose Foundry if you're a developer or enterprise building browser-based AI agents and need a high-fidelity, noiseless simulation environment for training and evaluation. They serve entirely different use cases; the decision hinges on whether your problem is restaurant operations or agent development.
GeneratedFrequently asked questions
Best-of guides
Used Foundry? Help shape our editorial sentiment research.