Durable runtime
Runs that finish — even when infrastructure doesn’t.
Every step is checkpointed. Retries resume from the last completed step, so a crash never means starting from scratch.
Cloud plans from $0 — bring your own model keysShip agents.
Ship agents.
Keep control.
Clara is the open agent runtime for teams that have outgrown demos. Build in code or Studio, then run with durability, isolation, and observability already wired in.
BYOK by defaultSelf-hostableSource-available
support-triage / production
live
Agent graph
Agent
sonnet-4.5
MCP · CRM
3 tools
Memory
tenant scoped
Policy
Max iterations10
Retries3 · exp
Run trace
run_01JY8K2P9Z
4.2s
total duration
00:00.0run.started
00:00.2memory.search12ms
00:00.5llm.started842ms
00:01.4tool.call · mcp.crm.lookup218ms
00:01.8tool.call · http.enrichretry 1/3
00:02.6llm.stream128 tok/s
00:04.2run.completed$0.018
9
events
2
tool calls
0
errors
Durable
checkpointed runs
MCP-native
tool execution
Isolated
Postgres RLS
Observable
live run streams
Flexible
any LLM via LiteLLM
Portable
cloud or self-hosted
One production surface
Your agent framework stops where Clara begins.
Keep the libraries you like. Clara gives every agent the operational layer it needs to survive production — without rebuilding your own control plane.
Tool execution
MCP, code, browser, and HTTP. One contract.
Give every agent exactly the capabilities it needs, with sandboxed execution and namespaced tool access built in.
Memory
Scoped context without rebuilding the storage layer.
Vector memory is tenant-isolated, namespace-aware, and available to agents through the same production runtime.
Operations
Control is part of the platform, not an afterthought.
Inspect live runs, stream events, audit changes, manage access, and attribute cost down to each tenant.
From spec to signal
One workflow. No operational gap.
Define the behavior, deploy a versioned spec, and inspect every event from one continuous system.
01
Define
Author typed agents with the SDK and CLI, or shape the same primitives visually in Studio.
›clara push support-triage
02
Run
Clara checkpoints every step and safely executes memory, MCP, code, browser, and HTTP tools.
›run.status active
03
Operate
Stream output, trace tool calls, audit changes, and understand cost across every tenant.
›trace 9 events · 4.2s
The agent engineering loop
Production agents are not shipped once. They are engineered in a loop.
Clara closes the gap between authoring and operations. Every run produces the trace you need to make the next version deliberate — not guesswork.
One agent specImmutable versionsFull run trace
Clara loop
Run evidence returns to the next version.
01
Define
Code or Studio
02
Version
Immutable spec
03
Run
Durable execution
04
Observe
Trace, cost, audit
05
Refine
Prompt, tools, policy
repeat with evidence
01
Define
Code or Studio
02
Version
Immutable spec
03
Run
Durable execution
04
Observe
Trace, cost, audit
05
Refine
Prompt, tools, policy
back to Define
Clara Studio
Visual when it helps. Code when it matters.
Studio is not a separate no-code toy. It reads and writes the same versioned agent spec as your SDK, so technical operators and developers stay on one production contract.
- Conversation-first agent building
- Live validation on every change
- Visual tools, triggers, models, and memory
- Immutable versions with full run traceability
Clara Studio · support-triage
valid specBuilder
Add CRM lookup, scoped memory, and retry failed HTTP calls.
I updated the agent graph and added an exponential retry policy. The spec is valid and ready to save.
+ mcp.crm.lookup
+ memory.search
+ retry: exponential
+ memory.search
+ retry: exponential
Canvas · v12
Support triage
agent
CRM lookup
mcp
Memory
vector
HTTP enrich
retry ×3
Built for platform teams
Give every tenant a hard boundary.
RLS by contract
Tenant isolation lives in Postgres policy, not in application convention.
Granular RBAC
Owner, admin, developer, and viewer roles share one exhaustive permission matrix.
Audit everything
Administrative actions, runtime events, and usage are attributable and inspectable.
Deploy your way
Use Clara Cloud or run the same platform in your own infrastructure.
Clara Cloud
Pricing built around how agents run.
Agents and seats are simple plan gates. Runs are the meter — predictable at the base, flexible as you scale. Your model cost stays yours with BYOK.
Included run quotas reset monthly. Usage over the included quota is billed progressively; high-volume contracts receive custom rates.
Free
$0USD
forever
Explore the platform and ship your first agents.
For prototypes & hobby projects
- 1,000 runs / month
- 3 agents
- 1 seat
- 7-day run history
Most popular
Pro
$49USD
per month
Everything a developer needs to run agents in production.
For solo developers & small teams
- 25,000 runs / month
- Unlimited agents
- 3 seats
- Sandbox included
- 30-day run history
- Studio — limited mode
Team
$299USD
per month
Production controls for teams operating critical workflows.
For serious production teams
- 250,000 runs / month
- 10 seats
- Full Studio
- Full RBAC matrix
- 90-day run history
- Audit log export
Enterprise
Custom
annual agreement
Isolation, governance, and support for regulated workloads.
For platform & security teams
- Dedicated database
- SSO / SAML
- Custom retention
- SLA & priority support
- Enterprise deployment options
Self-hosted Clara
Your infrastructure. The same control plane.
Start free for non-production use, then license each production deployment. Your data, model keys, and runtime stay inside your boundary.
Read the self-host guideCommunity
$0
Non-production under BUSL
Pro
$199/mo
Production + Studio
Business
$699/mo
SSO, audit, multi-environment
Enterprise
$30K+/yr
Dedicated DB, SLA, support
The agent control plane