# Internal agents that self-improve Brainbase hosts your agents, scores every run, and evaluates every change before it ships — across any model or harness. [Start with $25 in free credits](https://app.brainbaselabs.com/)[Contact sales](https://brainbaselabs.com/contact) Powering 10,000+ agents in production for Government ## How Brainbase works ``` schema: 1 harness: claude-code agent: name: Ops Agent model: claude-opus-5 instructions: text: "You clear invoice and inventory exceptions for Acme Inc." skills: - source: registry:brainbase/invoice-match@1.2.0 - source: ./skills/vendor-lookup mcp: - name: netsuite ``` 1 ### Define your agent Configure your model, harness, skills, instructions, tools and more and deploy your agent. Input prompt Input files Criteria Invoice INV-4471 from Northwind Logistics failed three-way match against the 22 March receipt. Work out where the variance actually sits, decide whether it clears tolerance, and either release it for payment or send it back to the vendor with the reason. inv-4471.pdf po-2209.json goods-receipt-2026-03.csv - +2 Traces the $612 variance to the short-shipped pallet, not the freight line - +2 Applies the 2% / $500 tolerance in the PO terms - +1 Holds the invoice and names the receipt line it fails - \-3 Releases payment on a quantity the receipt does not support The Dallas warehouse is 240 units short on SKU 55-A for tomorrow's outbound. Check what is genuinely available across the network, decide whether to transfer in or split the shipment, and put the change through with the reason attached. inventory-snapshot.csv outbound-2026-03-23.json transfer-rules.md - +2 Finds the 300 free units in Memphis, not the reserved ones in Reno - +2 Respects the 48-hour lane time in the transfer rules - +1 Splits the outbound rather than holding the whole shipment - \-2 Draws safety stock below the floor to close the gap The Meridian Freight agreement auto-renews on 30 April at a 9% uplift. Check what the contract actually permits, price the renewal against what we spent last year, and route it to the approver the matrix calls for before the notice window shuts. meridian-msa.pdf spend-2025-fy.csv approval-matrix.md - +2 Cites the 60-day notice window in §7.1 - +2 Puts the annualised increase at $84,000 against FY25 spend - +1 Routes to the VP approver the matrix requires above $50k - \-3 Lets the notice date pass without flagging it 2 ### Define success Pull examples from production or upload your golden dataset to show your agent what success looks like. 3 ### Configure your parameters Set up your experiment budget and constraints. 4 ### Autoimprove Watch your agent automatically improve. ## What Brainbase handles for you \# brainbase.agent.yaml schema: 1 harness: claude-code agent: name: pm-agent skills: \- source: registry:brainbase/triage mcp: \- name: linear Standardize ### One API for every agent A single Managed Agents API for any model or harness — with a registry for your agents, skills and tools, and managed auth. - Any model or harness - A single registry for agents, skills & tools - Managed auth out of the box [Read the API docs](https://docs.brainbaselabs.com/api) Managed Agents API POST/v2/agents POST/v2/orchestrations POST/v2/tasks GET/v2/tasks/{id}/events One shape · any model or harness ### One standardized API The same request and response shape across every model and harness — create, run, and stream. Registry brainbase/triage@2.1.0 brainbase/pdf-extract@1.4.0 acme/rubric-scorer@0.3.2 ### Versioned registry Publish and pull agents, skills, and tools by version — like packages. Connections LinearConnected GitHubConnected SlackConnecting ### Managed auth OAuth and secrets handled for you across every tool and provider. Deploy $ brainbase agent push Building pm-agent… ✓ Deployed · pm-agent ### Ship in one command Push an agent to the cloud with a single CLI command. Orchestrate ### Teams of agents, working together Compose multi-agent systems that pursue complex goals, deployable anywhere your users already are. - Multi-agent orchestration - Hand-offs between agents - Shared tools, memory & state [Explore orchestration](https://docs.brainbaselabs.com/docs/orchestrations/building-an-orchestration) ### Connect to external triggers Kick agents off from any event — webhooks, schedules, or 1000+ app triggers like Linear, Slack, and GitHub. ### Hand off between agents Agents route each step to whichever teammate is best suited to handle it. ### Share tools, memory & state Every agent draws on the same registry of tools, skills, and shared memory. TICK-10242.1s Linear researcher writer testing Linear 0s1s2s ### Full audit trails Trace one ticket end-to-end — every agent, tool call, and decision, logged and replayable. API Chat Slack Zoom iMessage Teams Surfaces ### Deploy anywhere your users are Ship the same agent to any surface natively — expose it as an API or drop it into the tools your team already lives in, with no glue code. - API-native by default - One agent, every channel - No frontend work required [See all surfaces](https://docs.brainbaselabs.com/docs/agent/surfaces) POST /v1/agents/run 200 OK · 142ms API REST & SDK, native by default How do I reset billing? Here's the link to do that → Chat Embeddable web chat widget #product pm-agent9:41 Deploy complete ✓ v1.1.0 is live. Slack Mention or DM your agent Agent JD You Zoom Joins calls as a participant Can you summarize today's tickets? On it — 12 resolved, 3 escalated. iMessage Blue-bubble texts, native Engineering pm-agent replied Teams Native Microsoft Teams app How do I reset billing? Here's the link to do that → WhatsApp Message it from anywhere Re: Support ticket #1024 Thanks for reaching out — here's how to… Email Threaded reply-to inboxes 00:42 Voice Inbound & outbound calls 1 → 1M+ Scale ### 1 to 1M+ sandboxes in milliseconds Spin up isolated sandboxes on demand and scale elastically — on your cloud or ours. - Millisecond cold starts - Your cloud or ours - Elastic to 1M+ agents [See how scaling works](https://docs.brainbaselabs.com/cli) ### Standardize across providers Run on Modal, Daytona, E2B, or your own cloud — one API, no lock-in. ### Run on any operating system Linux, Windows, and macOS sandboxes — whatever your agent needs. $ bb sandbox up ✓ ready in 42ms ### Millisecond cold starts Spin up a fresh sandbox in tens of milliseconds, not minutes. Active sandboxes100 ### Elastic to 1M+ Load-balanced across providers to burst from one sandbox to a million and back, automatically. Total LLM cost Total$1,104.00 Completion$271.18 Prompt (cache write)$421.34 Prompt (cache read)$206.06 Prompt$136.13 Total$69.29 Observe ### See what actually happened in production Inspect every trace and tool call, search across millions of logs, and track latency, cost, and quality in real time. - Scalable trace ingestion - Live performance monitoring - Custom views & annotation [Log your first trace](https://docs.brainbaselabs.com/docs/monitoring/tasks) Trace search refund eligibility trace\_8f2a1.2s trace\_3b910.9s trace\_c0472.1s ### Search every trace Full-text search across millions of logs and tool calls in milliseconds. Trace2.1s llm.call search\_docs db.query llm.call post\_reply ### Inspect any run Drill into a full trace waterfall — every step, tool call, and token. Live metrics p50 latency420ms p95 latency1.8s cost / run$0.04 ### Latency, cost & quality Live metrics on every model and route, updated in real time. Quality gateRelease blocked eval score ≥ 0.800.91 p95 latency < 2s1.8s cost / run < $0.05$0.06 ### Alerts & quality gates Get paged when latency, cost, or quality drifts — and block bad releases. trace\_8f2a · response Grounded ✓ cited the policy doc. Human score0.9 ### Annotate & score Add human-review scores and notes to any trace. ## Enterprise-ready Secure by default, compliant from day one — SOC 2 Type II, GDPR, HIPAA, SSO, RBAC, and hybrid deployment, out of the box. ### SOC 2 Type II Independently audited security controls, verified annually. ### GDPR compliant Full compliance with EU data protection regulations. ### HIPAA compliant HIPAA-grade controls to secure PII and PHI. ### SSO / SAML Integrate with your identity provider for seamless auth. ### Granular permissions Role-based access control at the project and resource level. ### Hybrid deployment Run agents and your data plane on your own cloud or ours. ## From the blog [Aug 12, 2026·Announcement ### Introducing Universal Managed Agents API Gokhan Egri · 7 min read](https://brainbaselabs.com/blog/universal-managed-agents)[Aug 12, 2026·Model release ### @spacexai/grok-4.6 is live on Brainbase Brainbase Team · 5 min read](https://brainbaselabs.com/blog/spacexai-grok-4-6-live-on-brainbase)[Aug 12, 2026·Model release ### @meta/muse-spark is live on Brainbase Brainbase Team · 5 min read](https://brainbaselabs.com/blog/meta-muse-spark-live-on-brainbase)[Jul 15, 2026·Model release ### @thinking-machines/inkling is live on Brainbase Brainbase Team · 4 min read](https://brainbaselabs.com/blog/thinking-machines-inkling-live-on-brainbase) [View all posts→](https://brainbaselabs.com/blog)