See every step your AI agents take.
Charter Bots is the control room for teams that run AI agents in production. Follow each run from question to answer — every tool call, every handoff between agents, and the time and cost behind it.
Initial design partner program · Free during the program · Small group, limited places
AI agents are easy to launch and hard to run.
Once agents start calling tools and handing work to other agents, a single answer can hide dozens of steps. When something is slow, wrong or expensive, nobody can say why.
Slow answers, no reason
A reply takes a minute and a half. Was it the model, a tool, or another agent it asked for help?
Costs you can't attribute
The bill grows every month, but it isn't tied to agents, conversations or the questions that drove it.
Silent failures
A tool fails, the agent retries and still answers — so the failure never shows up anywhere you look.
Everything an agent did, on one screen.
Charter Bots turns each agent run into a timeline you can read in seconds, and rolls every run up into the numbers your team needs.
Run timelines
A flame graph of every run: model calls, tool calls and handoffs, sized by time. Zoom into any step.
Agent-to-agent handoffs
When one agent asks another for help, the whole chain stays in one run, with each agent's own work nested inside.
Cost per run and per agent
Tokens and estimated cost for every call, every run and every agent, including how much of each prompt was served from cache.
Critical path
See the exact chain of steps that set how long an answer took — the steps worth making faster.
Tool errors, surfaced
Failed tool calls are flagged even when the run still completed, with the error message one click away.
A chat console that shows its work
Talk to any agent and watch each step as it happens — then open the full run behind any reply.
Bring your own tools. Build agents that stay in their lane.
Connect the APIs and tool servers you already run, then build focused agents that can use only what you give them.
Your APIs, as safe tools
Turn the APIs you already run into typed, named actions your agents can call — without handing agents your credentials.
- Each API becomes a set of described actions
- Credentials stay on the server side
- Read-only by default; you choose what to expose
Onboard your own MCP servers
Plug in any MCP-compatible tool server, hosted or your own, and choose exactly which of its tools each agent may use.
- Connect by URL with secret-backed authentication
- Tools are discovered automatically
- Write actions stay off unless you opt in
Agents with a limited scope
Build focused agents that can only use the tools, data and teammates you give them — each in its own sandbox.
- Allow-list exact tools and the agents it may ask
- Isolated runs with no direct access to credentials
- Pick the model and instructions per agent
- Optional human approval before sensitive actions
Ask an agent. See how it got the answer.
Every reply in the Charter Bots console comes with its working: what the agent thought, which tools it ran, which agents it asked, and what the answer cost.
- ✓Step by step, in order. Thoughts, tool calls and handoffs appear as they happen.
- ✓Open any step. Inputs, results and errors for each tool call.
- ✓Numbers on every reply. Duration, tokens and cost, with a link to the full run.
- ✓Every conversation kept. Search past chats and pick up where you left off.
Live in an afternoon.
Charter Bots reads the activity your agents already produce. No changes to your agents, and your data stays where it is.
Connect
Point Charter Bots at your agent environment with read-only access. It discovers your agents and the tools they use. Add your own APIs and tool servers when you're ready.
Watch runs arrive
Every conversation becomes a run with its steps, timing, cost and errors — including work handed between agents.
Act on it
Filter to failed runs, find the slow step, compare agents by cost, and chat with any agent to check a fix.
Built for the people who keep agents running.
Platform teams
One view of every agent, what it's allowed to use, and how it behaves in production.
On-call engineers
Go from "the bot is slow" to the exact failing step in a couple of clicks.
Agent builders
See how prompt and tool changes affect speed, cost and errors, run by run.
Engineering leads
Know what each agent costs and whether it's getting faster or slower over time.
We're choosing our first design partners.
Charter Bots is in its initial design partner phase. We're working closely with a small group of teams, free of charge, to shape the product before general availability. Team and Enterprise plans come later.
Design partner Open now
For a small number of early teams who want to shape the product.
- All features
- Up to 10 agents
- Bring your own APIs and MCP servers
- Direct line to the founders
- Input on the roadmap
Team Later
For teams running agents in production, after the design partner phase.
- Unlimited agents and tool servers
- Longer history
- Alerts on failures
- Team workspaces
Enterprise Later
For organisations with many agents and strict controls.
- Deploy in your own environment
- Private tool servers and APIs
- Priority support
- Single sign-on and roles (on the roadmap)
Questions, answered.
Do I need to change my agents?
No. Charter Bots reads the activity your agents already produce. There's nothing to add to your agents' code.
Where does my data go?
Charter Bots runs in your environment with read-only access. Conversations and results stay under your control.
Does it work with agents that call other agents?
Yes. Handoffs between agents are kept in one run, with each agent's own steps nested under the call that started them.
Can I connect my own APIs?
Yes. Each API is exposed to agents as a small set of named, described actions. You define the actions once and reuse them across agents; credentials stay on the server side and never reach the agent. There's no one-click import from an API spec yet.
Which tool servers can I onboard?
Any MCP-compatible server, connected by URL with secret-backed authentication. Its tools are discovered automatically, and you pick which ones each agent may use.
How do I limit what an agent can do?
Every agent gets an explicit allow-list: which tools, from which servers, and which other agents it may ask for help. Agents run isolated, you choose their model and instructions, and sensitive actions can require a person's approval first. Anything not on the list is out of reach. Per-agent spend limits aren't available yet.
How is cost calculated?
From the tokens each model call used — input, output, and prompt cache reads and writes — at the model's published rates.
What does a design partnership involve?
Free access to everything during the initial design partner phase, a short call every couple of weeks, and your feedback shaping what we build next. We're keeping the group small, so we review each application.
Stop guessing what your agents are doing.
We're taking on a small group of initial design partners. Leave your work email to apply, and we'll reach out if it's a fit.