More Agents
Kimi Code
Quick answer
Prism writes a
[providers.prism] block into Kimi Code's config.toml, adds a [providers.prism-responses] block when you have Responses models, and creates one [models."prism-…"] alias per model with its capabilities and reasoning efforts.Last updated Reviewed against Prism v0.3.26
Config locations
| File | Path |
|---|---|
| Providers and model aliases | $KIMI_CODE_HOME/config.toml, or ~/.kimi-code/config.toml |
| MCP servers | $KIMI_CODE_HOME/mcp.json, or ~/.kimi-code/mcp.json |
Kimi splits the two concerns into separate files, so setting a provider never risks your MCP entries and vice versa.
One-click setup
1
Open the Agents tab
2
Click Setup
Prism strips its previous managed region and writes a fresh one, keeping a one-time backup of the original file.
3
Choose a Prism model
The aliases appear in Kimi's model list with their display names and limits.
What Prism writes
[providers.prism] type = "openai" base_url = "http://127.0.0.1:11434/v1" api_key = "prism" [providers.prism-responses] type = "openai" base_url = "http://127.0.0.1:11434/v1" api_key = "prism" [models."prism-glm-5-1-cloud"] provider = "prism" model = "glm-5.1:cloud" max_context_size = 128000 display_name = "GLM 5.1 (Ollama Cloud)" max_output_size = 16384 support_efforts = ["low", "medium", "high"] default_effort = "high" capabilities = ["tool_use", "image_in", "thinking"]
How the alias keys are built
The alias key is the model's route key lowercased, with every character that is not a letter or digit replaced by a hyphen, prefixed with prism-. That keeps the key safe for TOML quoting while the model field still carries the exact name Prism expects.
Provider is chosen per model
Each alias points at
prism or prism-responses according to that model's protocol. The prism-responses block is only written when at least one Responses model exists.Capabilities and reasoning
| Field | Value |
|---|---|
| capabilities | Always includes tool_use; adds image_in for vision models and thinking for reasoning models. |
| support_efforts | The model's configured reasoning efforts, defaulting to low, medium, high. |
| default_effort | The highest of the supported efforts, so the model starts at full strength. |
Add MCP servers
Kimi Code reads MCP servers from mcp.json. Enabling Kimi Code in the MCP tab writes an entry there — a plain url with no transport field means HTTP. See MCP Gateway.